Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium IVIember Libraries http://www.archive.org/details/reputationeffectOObane 531 1415 fl-jq working paper department of economics Reputation Effects and the Limits of Contracting: A Study of ttie Indian Software Industry Abhijit Banerjee Esther Duflo No. 99-14 July 1999 massachusetts institute of technology 50 memorial drive Cambridge, mass. 02139 WORKING PAPER DEPARTMENT OF ECONOMICS Reputation Effects and the Limits of Contracting: A Study of the Indian Software Industry Abhijit Banerjee Esther Duflo No. 99-14 July 1999 MASSACHUSEHS INSTITUTE OF TECHNOLOGY 50 MEMORIAL DRIVE CAMBRIDGE, MASS. 02142 Reputation Effects and the Limits of Contracting: A Study of the Indian Software Industry Abhijit V. Banerjee* and Esther Duflo^ First Draft: November, 1998 - This version June 11, 1999* Abstract This paper examines evidence of the role that reputation outcomes. We We plaj'S in determining contractual conduct an empirical analysis of the Indian customized software industry. analyze a data set containing detailed information about 230 projects carried out by 125 software firms that reputation matters. varj' Ex we had previously collected. The evidence supports ante contracts as well as the outcome after cx-post renegotiation with firms' characteristics plausibly associated with reputation. pattern is the view that We argue that this not consistent with optimal risk sharing and propose a model of the industry where reputation determines contractual outcomes, whose predictions are consistent with several facts observed in the data. We argue that there is no obvious alternative explanation to the patterns present in the data. 'Massachusetts Institute of Technology. Email: banerjce@mit.edu * Massachusetts Institute of Technology. Email: edufio@mit.cciu *Wc thank Daron Acemoglu, PhiHppc Aghion, Roland Benabou, Mathias Dewatripont, David Gcncsove, Jonathan Guryan, Oliver Hart, Doug Jean Tirolc, and Richard Zcckhauscr We Miller, Dilip Mookhcrjec, Ashok Rai, for helpful conversations Emmanuel and Paul Joskow for encouragement and support. acknowledge financial support from the the National Science Foundation, Alfred Shultz Fund and the John D. and Catherine professionals in India, who Mac Arthur patiently answered our Foundation. many We Saez, Andrei Shleifer, P. Sloan Foundation, the are especially grateful to the software questions, and in particular to Mr N. R. Narayana Murthy, from Infosys Technologies Limited, whose help and insights were crucial throughout this project. 1 Introduction The idea that there arc severe hinits to what can be achieved througli contracting has had an enormous impact on the way economists now tliink about Correspondingly, there has been a growing emphasis on firms, marl<ets and governments. of reputation as a tlie role way of coun- teracting the problems created by the hmitations of contracting.^ Wliile less often emphasized, a view of the world which gives central importance to issues of contracting, reputation and trust, also has portantly, it important consequences for the process of growth and development. Most im- suggests that the lack of a proper infrastructure for contract enforcement (which makes contracting less effective) and the important determinants of success human determinants such as difficulty of building a secure reputation^ are potentially in getting out of poverty, along with the more conventional and physical infrastructure. capital This paper attempts to quantitatively assess the importance of reputation and, by implication, the seriousness of the limits software industry. sired way on contracting Customized software is in the context of the Indian an obvious place to study such end-product tends to be extremely complex and the contract themselves do not fully understand process. Therefore, it ahead of time in a In fact, typically the parties to what they want until well into the production seems naive to expect that they could write a contract enforceable by the courts that would fully cover all contingencies that could arise in the production process. More- over, software production docs not require very simply own a number of PCs (which much are cheap fixed capital: indeed most firms nowadays and getting cheaper). The premises, access to a mainframe and links to a satellite, can sibility of effects since the de- difficult to describe that a third party (such as a court) would understand. customized all be rented."^ rest, including the This limits the pos- the reputation effects that interest us being confounded with the effects of deferential access to capital or the lack of real competition. The Indian is software industry an industry which is is suitable for such a study for a hi Lidia, the of reasons: First, prejudice, or a history of Murphy bad performance, government has actually invested heavily, (1997). as and by emphasized by Tirolc (1996). all accounts fruitfully, to make sure that firms have the option of renting expensive fixed inputs (such as expensive computers, building space and equipment satellite it quite large (employing 140,000 people with a turnover of $1.75 biUion in 'Sec for example Grcif (1994), Baker, Gibbons and Stemming from number telecommunication) in virtual "Software Techonology Parks" (STPI, 1997). for 1997-98) and growing fast (at an average annual growth rate of Second, its large (over main focus is on exports (more than of 30%) and fast-growing share of the exports industry's current focus is revenue comes from exports) and a its is over the past six years). customized software. Moreover, the on expanding the export of customized software relative to businesses on the grounds that this 1997).'' 60% 54% is be likely to its best bet for the near future other (NASSCOM, Consequently, the limits of contracting are a major issue in this industry and one that everyone is clearly concerned about. Finally, the fact that the contracts are typically across long distances makes contracting and, perhaps more complicated more importantly, because Ijoth by making monitoring somewhat harder of the inherent difficulties of international litigation (combined with the deficiencies of the Indian court system). comes from interviews of 125 software companies in India (Bangalore, on the two a total of Prima how major software development centers collected detailed data and the software buyer - supposed to pay is for all realized costs. henceforth, the client fraction of the contracts - is very limited and the view and the client quarter of the time and material contracts. There is post: the pays less and 2. Figure 'in the past, export of software services 1 - is - henceforth, the firm and material contracts for all realized costs. A large buyer docs not pay the entire than the full amount also a simple pattern in contract that gets chosen and the sharing of the costs which in Figures 1 In time supposed to pay do however get renegotiated ex cost in almost half the fixed contracts shown initially sample are either fixed-price contracts or All contracts in our important. is time and material contracts. In fixed-price contracts the software firm gets a fixed price on the company and set). the data supports both the view that contracting is use in this paper the contract got renegotiated as the project evolved (wc have 236 contracts in our data that reputation The data we they have completed, including what kinds of contracts were last projects facie, in three Wc Hyderabad and Pune)^. arranged between them and is its in about a both the kind of a result of the renegotiation. It shows the fraction of fixed-price projects as a function was almost exclusively on-site services (or "body-shopping"), cf. Hecks (1996). ''In We each city, wc interviewed half of the firms who belong selected the firms randomly, but we oversampled the refused to meet with us and answer the questionnaire. CEO's unavailability at the moment we were to the software technology park (all exporters do). firms that are not fully-owned subsidiaries. Some appointments No firm could not be arranged due to the interviewing, and these firms were replaced. and Figure of the foundation date of the software firm, 2 shows tlie share of overrun'' paid for by the firm as a function of the age of the software firniJ Botli arc sharj^ly increasing with the starting date of the firm. In particular, firms created in 1994 or after (half of the sample) Ijcar a substantially larger share of the overrun than older firms they bear is more sharply with age over increasing on average, and the share of overrun this range. Measured both in terms of the ex ante contract and in terms of ex post outcome, young firms bear a larger share of the 'risk' of each software project. This effect of effect of and have age perhaps the main empirical finding of the paper. is We interpret this as an reputation on the grounds that the firms that started in the industry a long time ago sur\'ivcd arc more be the kinds of firms that clients can trust likely to - the older firms that cannot be trusted arc likely to have already gone out of business (since eventually people would have got to know about them). To provide in Section 5 further support for our interpretation of the age effect as a reputation effect, we show that a similar pattern exists when we use other potential measures of reputation such as whether there has been a previous transaction between the firm and the client, whether it is an internal project term arrangement with the firm)^, to some extent among with a client firms that work an internal for industry ^^hich explains basic idea of the know who was why model reliable: reliable firms '"The is that this amount is is between young and old firms disappears we develop a simple model based on our observation of the and the that in most cases by the end of the project the firm it if While this is client not contractible, firms and clients they could commit to always follow a certain norm. clearly there can be other norms that will also work - is The that of being always try very hard to ensure that they do not exceed the cost overrun when they fail typically not consistent with short-run profit maximization of the project cost that goes Because the number of firms per year all or has a long- that different kinds of reputation are that they had implicitly promised, and pay for any extra overrun problem owns the firm reputation would have the observed effect on contractual outcomes. could nevertheless benefit from - either client. responsible for cost overruns. norm we emphasize here who we show Further, substitutes. For example, the difference In sections 3 and 4 of the paper The etc. (i.e., beyond the in the sample these firms together. ''Wc will describe this type of structure below. to do so. The by the firm or initial prediction. is small for firms created before 1988, wc have grouped the client, and can onty be sustained convcntionall}', value on reliable who tlicir We reputation. rest are not them look at equilibria where a certain fraction of firms and clients are and investigate the implications of what happens buyers from unreliable more there should be smaller, which is a change in the fraction of those fixed-price contracts if of other predictions from this who we describe above. are likely to be reliable We also argue that a is we cannot provide sharp enough evidence The objective of this paper is However our reputation-based story does it rules out where there Of number it clear is to rule out alternative reputation models: possible, for example, that the relevant reputation reliability. is model are consistent with what we observe. While we do provide some evidence supporting the broad premises of our model, that basic true of time and material contracts. Therefore is the share of buyers consistent with the evidence is The that fixed-price contracts arc best for protecting reliable while the reverse sellers some to put arc reliable (interpreted as a change in the average reputation of the firms). trade-off that governs more the firms and chcnts arc cither innately reliable or, the particular equilibrium that they arc playing induces if and the if is for not to distinguish rule out many it honesty or for a different form of among different kinds of reputation. alternative explanations. In particular models where there are no agency problems as well as models of agency problems is no learning about the firm's type. course, this is all conditional on establishing that data when we impose the reputation model on what we arc picking up here is the effect of it. 6, are in fact correctly interpreting the In other words, some other with these measures of reputation. In Section we it still variable which we consider some remains possible that happens to be correlated They of these explanations. broadly in two classes. fall First, there is a class of alternative explanations which rule out agency problems: the differ- ences in the contracts is then explained cither by differences in risk-sharing or by differences in the production technology available to the firm. Against this view, implausible that the contractual variations that The basic point is that in our data set, firms are usually young firms are especially small. of the risk we observe It is (57% on average) and why we first argue that it is very are a result of optimal risk-sharing. much smaller than therefore very hard to understand especially the smallest why their clients firms bear so and much and youngest firms bear the most.^ In response to the view that there arc differences in the production technology (essentially that ''There arc of course other determinants of the sharing of the risk. We discuss these issues in section 6. young firms arc more incompetent) wc point out that the natural should be to lower the price the j'oung firms gets paid rather than they can lisk that ferences in to make them Moreover the evidence docs not support afl'ord. ill competence between the firms is of a such incompetence effect of bear a lot view that the tlie of dif- magnitude that can explain the differences in the contracts. For example, \vc present in Figure 3 the average overrun as a function of firm's foundation date. pay If the high shares of overrun paid by young firms were a we should for higher overrun, seems to be class of competing theories posits that there are agency problems but no learning with the evidence on the effects of sources of this conflicts reputation other than age. These and related issues are discussed at firms. a final piece of evidence, is recognized and we emphasize the some length and at the industry level as well as by individual For example, the National Association of Software Services Companies (NASSCOM) The main element they is emphasized repeatedly stress in the process of acquiring it is the number of Indian firms that have is a way association provides technical consulting to any The Indian government level, efforts to This paper ISO 9000 1997). certification or are for firms to establish member who wants provides financial incentives for firms who a reputation). to get acquire it. ISO The certification. At the individual develop a reputation arc also obvious.^" a part of a small but growing is contractual choice. ^^ to this work. (NASSCOM, (ISO-certified firms have proven that their software development processes follow approved routines, which is in Section 6. fact that the necessity to build reputation directory of the Indian Software industry has a large section on "quality" more make them see average overrun falling with age. If anything, the opposite about the firm: we point out that trust to true. The second As way Among They examine number of papers that study the empirics of recent papers Crocker and Reynolds (1993) is most closely related the determinants of the choice between fixed-price contracts and flexible contracts in U.S. Air Force engine procurement. In their view, the key trade-off the following: fixed-price contracts protect the government against ex post opportunism (in particular it makes '"20% of the firms in it useless for the contractors to claim higher costs) but they require the our sample already have ISO certification. "Monteverde and Tcece (19S2), early papers on this subject. specific investments 13% arc in the process of getting These papers differ from ours in studying settings where there arc huge relationship- and very long term relationships arc the norm and where the key trade-off of contract (or control) and it. Masten and Crocker (1985), Joskow (1987) and Pittman (1991) arc important flexibility. is between tightness an exhaustive ability to draft costly. Time and material room the list is and if contractor.-'^ Contracts will tend to by the for opportunistic Ijehavior short), them easy the contractor is to draft more (if likely to the engine well is known be fixed-price if or the production and two contractors confirms these Their work shares therefore a central intuition with ours: the reputation of the predictions. contractor does matter for the choice of contracts. ^'^ is possible, but behave opportunistically. Their empirical analysis of a panel of 44 contracts between the government it is contracts do not require a truly complete agreement ex ante, but open the nature of the engine makes cycle of requirements (a complete contract), which that the contract will be fixed-price The . The more reputed a central difference is firm is, the less likely that fLxcd-pricc contracts are not associated with any ex post cost for the contractor, since fixed-price contracts are "truly complete agreements". contract is In contrast, we recognize the fact that in the software industry the never complete. Fixed-price contracts need not be more precisely drafted than time and material contracts. Overrun happens between containing opportunism by the Lafontaine and Shaw (1996) is in both types of contracts. The central trade-off client and opportunism by the another paper that looks at the is firm. efi'cct on of a firm's age contracts (in the context of franchising) and finds that the franchiser's age has no effect on the However contract. as they point out, by changing the franchise contract over time a franchiser runs the risk of hurting its early franchisees (who are locked into one contract while their competitors get a different contract that perhaps allows them to be more aggressive). Because of this cross-contract externality, contracts may not change very market over time becomes more knowledgeable about the The rest of the paper is organized as follows: much over time, even if the franchiser.-''* In Section 2, we describe the institutional '^Bajari and Tadclis (1999) emphasize a related trade-off in the private sector building industry (in a model where there arc no unobserved difference among contractors). Time and material contracts give the contractor little incentive to control the costs, but do not require complete drafts. Fixed price contracts give the contractor strong incentives to control cost, but require more precise design to avoid costly bargainig during the completion of the project. contract is The time scale and the complexity of the project if changes arc needed will determine which chosen. ''in their paper, they measure the reputation by the number of litigation conflicts that the contractors had in the past. '""This paper is also related to papers such as Barron and limbeck (1984), Shcpard (1993) and Genesovc (1993) which test the implications of theories based on aymmetric information in industrial contexts (but not the implications for the choice of the contract). more settings in The model software. and present a number of basic detail presented in Section 3 and is facts its about the production of customized predictions about how reputation shapes the contractual forms as well as the ex post outcome arc described in Section In Section 4. 5, we provide evidence which, in our view, clearly supports the implications of this model. In Section 6, we discuss alternative explanations of the pattern observed in the data. Section 7 concludes. and Basic Facts Institutions 2 We begin Ijy describing the sequence of events leading to the off-shore production of a piece of The software.'^ when project begins the client sends a request for proposal to one or more firms. interested firm studies the request (this costs the firm 1.25% of the total project cost for Each among the median external project^*^), and submits a proposal, which includes, proposed mode of payment and an estimate of how much the chooses a firm, and the firm and the client agree on a contract. client mode estimate of effort needed to complete the project, a and etc.) client would have The The of specifications. starts. The - first - corresponding to that will be reached phase For the median project, it takes 10% project effort to complete this phase. ^^ At the end of this part of the project, of deliverables it is would cost When this of the a specified milestone happens the request changes. ''Tabic 1 client The is 'For the writing work is is client the lower level design, coding and testing of the software. reached, the firm sends the deliverable to the client. statistics client what the clarified. it mentioned does not own in this Each time has been delivered (by signing firms also send regular status reports to the clients (a shows the descriptive of the total usually clearer to both the client and the firm and the schedule can either acknowledge that "'Those projects where the which is sometimes amended or The second phase is firm, in collaboration with the user at the client's end, writes the set of functions that the software will execute. wants and what The of payment, financial details (price, phases of the software development process or to modules of the software completing the project). The work then to pay. contract specifics an a projected schedule for deliverables (which arc specific milestones in the course of other things, a little less off) or than once paragraph. the firm or does not effectively control the part of the firm working towards the completion of the project (see below). some projects, specifications writing and subsequent work arc decoupled. One firm - or the - writes the specifications, and another firm completes the project. client itself a In terms of project is clients up-to-date aljout the jH-ogress of the project. week on average), keeping the outcomes our main focus the difference between the amount estimate. entire chapter A be on overrun: overrun in industry of effort actuallj' needed to complete the project estimated effort given in the contract. mean by an will It is therefore important to be clear about parlance and the what firms standard textbook on software management (Pressman (1997), has an on estimation. He describes the process as follows: 'The project planner begins with a bounded statement of software scope and from this statement attempts to decompose software into problem functions that can each be estimated individually. Line of Code or function points (the estimation variable) then estimated is component for may each function. Alternatively, the planner choose another such as classes or objects, changes or business processes im- for sizing, pacted. Baseline productivity metrics line of (i.e., code per person- month or function point per person months) are then applied to the appropriate estimation variable and cost or effort for the function an overall estimate Our is (Pressman (1997)) for the entire project.' interpretation of this and other material in this what we have learned from industry sources) how much effort will is is correct therefore clearly not will actually take. book (which that the estimate is is also consistent with is the firm's best guess about be needed to complete the project, assuming that the firm's current under- standing of the project estimate combined to produce derived. Function estimates are This is and that meant the firm adheres to its own to be an unbiased estimate of important because it tells productivity how much norms}^ The effort the project us that overrun represents the extent of deviation from the firm's initial plan of action. This also tells us that overruns ought to be quite client are typically not very clear at the very himself. Moreover the client may common: first, because the needs of the beginning of a relationship - even to the client not put enough effort into understanding and explaining what he wants. Not surprisingly then, the firm often does not understand what the client really wants. When, to be in the course of the project, the made and needs of the client eventually become these are costly. Second, the amount of time and "In other words, the presumption behind the estimate wants and that the firm implements the project at its is level of productivity. changes have needed to design and that the firm has understood perfectly normal 8 effort clear, what the client code a piece of software evaluate ex ante, even difficult to is when the set of functions well- is defined (both for the client and for the firm), and will depend on the type of technology Ijeing used, the ability staff and the and the experience of the clearest of goals there is and one would expect that some projects to maximum 74% will firms try their hardest to control costs 24% of the initial estimate, overrun in the sample of the overrun is is to. initial negative overrun, and the Both firms and by the changes cause 48% in client. mean overrun themselves is 34%, and the of the overrun on average). ambiguity in the specifications clients are, of course, from being wasteful The average overrun varies a lot (its standard deviation 8% (the most frequent one being the loss of the project to delays occasioned it 250%). According to the firms, overruns are due mostly to client (these due to and did not understand what the client really wanted), 13% and and delays end up costing much more than they ought of the projects are completed with a positive overrun. changes required by the 20% all arises shows evidence from our interviews confirming that overruns arc indeed common: 1 turns out that amounts some unexpected problem also the risk that delays or destroys the project. Finally, not Table two companies. Third, with the best of staff of the (in so (i.e., manager Very few projects (less in the where the firm to cases due to internal is Another difficulties in the firm middle of the way) and than 5%) are completed with a clearly not zero. is aware of the possibility of much as they could have overruns.^'-' Overruns, apart been avoided by both parties being more diligent), lead to delays which are costly^" and are a potential source of conflict between the client and the firm Vertical integration (conflicts arise when each and contracts arc two ways of limiting the waste due foreign companies have set up 100% owned subsidiaries in India.'^^ export oriented, and carry out work for their mother clients as well. ' side blames the other for the overrun). A number to overrun. Many These subsidiaries are 100% company and in some cases, for other of Indian software firms have also entered into arrangements under For example, the template of a firm's contract specifies that "the effort estimates provided for the conversion and testing phases of this project have been provided by the software firm on a best estimate basis. If the scope of the effort changes as a result of discussions during the detailed design phase, the software firm will analyze the impact of changes on the project and may present revised schedules and costs. Changes in schedules and costs resulting from such changes will be reflected by an '"Delays, while rarer than overrun, are far from and in 25% amendment uncommon to this contract." in our sample: there are delays in of the cases where there was an overrun. '•"Including AT.S;T, IBM, Microsoft, INTEL, ORACLE, Fujitsu and Motorola. 19% of the cases, which the firm dedicates a part of This client. is what an "Off-shore Software Development Center" (OSDC). The called is employees, office space, and computers to a single foreign its sends a steady fraction of his software development needs to the firm, and of the facilities devoted to him. This making use OSDC becomes virtually a unit of the and the interests of the firm there much is we Since is less (i.e., fixed-price contracts, a fixed price is These contracts are by far the Under mixed contracts the The and while there may be overrun, entire product is is fixed for the specification phase only at the beginning of the is fixed only when specifications are written A striking fact is that there appears to be no "intermediate" contracts where the client and the firm agree on sharing the costs. While these contracts predict extreme outcomes This is this. It evident from Table 2, all may be more may be happy to do whatever Or their variants: in is terms of cost-sharing, we actually do not for by the firm and the or nothing of the overrun for the three types of contracts. overrun of the cases) while firms with time For example, the client in ^"^ which shows the fraction of overrun paid in fixed-price contracts, the actual ""Indeed there For example, there arc no turns out that a large fraction of contracts get renegotiated ex post. proportion of firms that pay 46% paid for on a time and paid for on a time and material basis. These contracts are the least frequent contracts belong to one of these three categories. ^"^ always observe is and not necessarily the case. Under time and material contracts, the external contracts (15%). all nor for agreed upon up-front, before the specification analysis. Typically in such cases the requirement analysis is OSDC observe the following types of external contracts: Under price for the complete project known. contracts: In such cases the most frequent: 58% of external contracts are fixed-price contracts. price material basis, though this among works regularly. contracts that are performed neither within We is it type of vertical integration: the are interested in contracts rather than vertical integration our focus in this paper the mother companies of the firm). more whom responsible for is ^^ the overrun should be wasteful. mainly on external contracts process. in effect a client are clearly better aligned why reason client for is client is often shared between the client and the firm and material contracts sometimes pay overrun in such cases than in Even for overrun (in (in general precisely because overrun entails less waste. not need to be very precise about what he wants since he knows that the firm will asked of it. some cases property Such contracts arc observed, albeit rarely, rights in the among product substitutes for cash payments. the procurement contracts for airplane engines studied by Crocker and Reynolds (1993). 10 22% of the cases). However is it from the figures also clear Table 2 that the in initial contract has an clear influence on which party bears the risk of the project: in fixed-price contracts, firms bear on average 15.5% of the overrun, while they bear on average 63% time and material contracts. in 51.5% in mixed contracts and Since fLxed-pricc contracts dominate our sample, this evidence also implies that firms bear a lion's share of the overrun (57% on average for external 76% projects, for the median external somewhat clients, this is at least project). Since firms arc typically a firm faces a fixed-price contract has the option of walking off the job. done, but may it will also avoid the may have some it If it for the pervasiveness of renegotiation. docs, there is arbitrage court (even though in generally, Finally, firms is no project in where the extremely make some and therefore we ought and clients CEO may is by internal who was responsible for difficulties. what is 'it was our at least it (3)), or is We with the client). More for case: as In is of their own making, had a number of conversations and we paid fault responsible for the overrun. it'. We also have some mentioned above, we asked what follows, we assume that described as changes due to ambiguities and overrun caused entirely client due to the due partly to both (column regression of the share of overrun paid all conflict renegotiation. reliable. sometimes the Changes required by the by the firm when the overrun some to expect are taken to be caused by the client's responsibility. In is concessions in order to avoid voluntarily pay for any overrun that of the firm told us firms questions about OLS cff"cct and going to court inefficient some occasions the firm reports a indirect evidence that this (column has already it our sample where the firm and the client went to because they care about their reputation for being the firm usually from our conversations with industry people we have the impression that people go to court very rarely more it overrun and, at least at early stages of the job, the second very costly. Firms and clients will therefore prefer to fact, paid for work that even First, bargaining power because will not get it dominate. Second, the court system in India going to court. In smaller than their surprising. There arc several potential explanations when much (2)). and delays coming from the client's side Table 3 shows the share of overrun paid client (column In column (4), (1)), entirely we due to the firm present the coefficient of an by the firm on the share of overrun which it caused. types of contracts, firms always pay more of the overruns entirely caused by their mistakes compared to the overrun entirely due to the share of overrun paid by the firm lies client. Moreover, in all cases but one, the inbetween these two numbers when the overrun 11 own is partly caused by each side. OLS Furthermore, the regressions indicate that, regardless of the initial contract, the larger the fraction of the overrun that a firm has caused, the larger has to pay (if a firm causes percent more of one additional percent of the overrun, it share tlie it bears approximately 0.20 it). In the next section we present a model of the industry which is based on the picture that emerges from the above discussion. The main elements we wish to capture in our model arc the following: ^to- the high levels of overrun, • • the fact that both sides arc responsible for overrun, • the use of simple ex ante contracts, • the fact that the contracts get renegotiated ex post, the fact that the ex ante contract continues to influence the renegotiated outcome, • • the fact that firms and clients care aljout their reputation for voluntarily pay for overrun that A 3 Model is of their being reliable and will often own making. of the Software Industry The model we propose in this section is an attempt to capture in as simple a way as possible what, on the basis of our experience in the industry, we see as the fundamental structures and conflicts in the Indian predicted by the model in the data. customized software industry. will, as wc shall sec, contracting outcomes that will be match up reasonably However, one could come up with other models, or at models, which also explain the data. 5. The In the end, however, it Wc will discuss some well with least what is observed combinations of other alternative explanations in Section remains plausible that elements of these other models could also be a part of any comprehensive story of the software industry in combination, perhaps, with the story we tell. In this sense, the model The premise main of the model is is meant to be illustrative rather than definitive. that software projects arc prone to cost overruns and that the conflicts arc over the apportioning of these cost overruns. Overruns can happen for two reasons. First, the chcnt could have been insufficiently diligent in deUneating his requirements 12 or he could have made realize that this is As a a mistake. not what he wants and The issue also happen because the firm was is whether the client will demands changes. The possibility that each side will and specifically the courts, this is be willing to compensate it either lazy or unlucky in the come from Since the overrun could cither side, blame the other for when firm if it is is, a product he might to adequately compensated. enough. Second, overruns could way there is it carried out the project. an overrun, there This need not be a problem it. happy of course, if can observe who was really responsible. Our assumption is a real outsiders, will Ije that not possible in most cases. This is clearly something of a caricature of reality: firms systems to ensure that it of defining deliverables and having the Once on a client signs off up clear, ex post, is who was to that point the firm presumably ascribable to the is had done what and blame to client sign off a deliverable, he scope for future disagreements. is comes up with firm the changes - since they arc Parcto improving - but only make least when the result, clients clearly it set up any overrun. The procedure for on each deliverable is committed to a large extent do try to one such system. to admit that was supposed to do. This clearly at limits the Nevertheless, there seem to be lots of disagreements and this fact that even after many milestones have been reached, there remains substantial ambiguity about what exactly needs to be done. Disagreements, Overrun and Contracts 3.1 We The capture the possibility of this kind of disagreement as follows. (F) to build a piece of software that will be worth on both sides throughout, so this costs). In is V to the client money best thought of as a a world where the client can describe the product it However we assume that = (C) wants the firm (we will assume risk-neutrality payoff and the costs as money wants perfectly and the firm also understands this description perfectly, the project should cost y adopt the normalization that y client (i.e., the estimate is y). We 0. in every project the actual cost will be positive, i.e., there will be overrun: the client's description of the project will always be incomplete and the firm will never understand it perfectly. Total overrun will thus and overrun caused by the client (yc)- Assume be the sum of overrun caused by the firm (y^ that firm. 13 all overrun is initially paid for the by the The amounts yp and yc, arc determined, of the overrun, by the firm and the chcnt in describing and understanding the project. that firms face a choice between a high level of yp, yp^ and a low level, faces a choice between yc and overrun may be - this cfTort S])ccifically put because controlling overrun takes wc assume Likewise, the client y£_. it because the firm (or the effort or has generated (say the firm lying about is client) its costs). The extra private benefits to the client and the firm of a high level of overrun arc, respectively, and Bp. Assume that yc— in Ceteris paribus, both firms and clients prefer high levels of yc_- gets to keep a part of the overrun by the respectively, yc> Be and yp— yF> Bp so that always it is efficient to Bq minimize overrun. However, we assume that both yc and yp arc known only to the firm and will + third parties such as the courts only observe total overrun {yc all yp)?''^ Moreover, we restrict contracts to being linear and in addition require that they do not involve throwing money. In other words, we only consider contracts where the P+ (1 — s){yp +yc) borne by the firm - (s where G P [0,1]) a prc-spccificd fixed is client payment and s is the share of the overrun and material and the contract with contract).'^'' It will s = = 1 else. Of (corresponding (which corresponds roughly to a time be key to our analysis that neither of these contracts obviously dominates the other. The fundamental trade-off comes from the is anybody to particular interest to us will be two extreme contracts: the contract with s to a fixed-pricc contract) away any pays the firm an amount and neither party makes any payments - its client: fact that s is one number that being used to give incentives to both parties: a high s will give good incentives to the firm but not to the client, while a low s contract does the reverse. As a result, be possible to implement the first bcst.^'^ To complete the description contract. will typically not it of the contracting process Our reading of the industry practice is we need that the contract is to say who usually proposed by the ^'In other words, the fact that, say, two people were assigned to the project for 14 weeks what they were actually doing "''The correspondence realized costs rather is - proposes the is verifiable but not they could have been really working on a different project for most of that time. not exact because a time and material contract typically pays a than a fixed payoff. markup on the Similar results hold for that case but the exposition is somewhat more cumbersome. -'"'This basic tension is very general. we have chosen here, can be relaxed if impossibility of implementing the first Our restriction to linear contracts, while vital in the discrete case arc prepared to go to the best in that case is 14 model where overrun varies continuously: a consequence of the results in Holmstrom (1982). we the firm. W^c will hold if make we allowed the 3.2 Norms When the outcome first maintained assumption, noting however that similar results would this our client to propose the contract. best cannot be achieved by contractual means, the behavior of the firms and the clients if is at least partly assume that there are two types of firms and two types of one type of client observes a norm of being reliable, possible to improve on the it is clients. norm-governed. Specifically Of these, one type of firm and by which we mean that they always pay for any overrun that they themselves have generated, as long as the other side does the same. In particular, they will pay for the that the other side pays for However, to tr}' if When if the contract stipulates words they do not necessarily maximize current in other the other side docs not act reliably they do not act reliably either maximize current maximize it - overrun that they have generated, even profits. Assume by contrast, unreliable firms and clients - profits. they simply always act to their current profits. a firm and a client are matched, they do not directly observe each other's types. Rather they assign probabilities to the other party being know about them. The reliable, on the basis of what they probability that the firm puts on the client's being reliable, 6c, and the probability that the client puts on the firm's being likewise. Op, therefore summarize their respective reputations. and We will return to the question of how these reputations are sustained their evolution later in this section. Contracts for Protection 3.3 In this setting, since the reliable firms and clients arc going to be sclf-rcgulated, the function of the contract versa. The is to protect reliable clients against contract protects because it provides a fall-back option unreliable. If a firm, for example, generates a large it opportunism by unreliable firms and vice amount when of overrun the other party and can be taken to court and forced to pay at least the share of the overrun refuses to it is pay being for it, has contracted to bear. However as we have already argued in Section 2, we expect these contracts to be renegotiated. We assume that the renegotiated outcome is a sharing of the overrun which is potentially different 15 from what when in the contract is tliere is a dispute, would also cxjject that and is represented by the share of the overrun paid for by the firm s*{s,9c,9p). more reputed natural to assume that s* will increasing in is We s. more bargaining power and therefore pay parties will have example, the court less (because, for It is pay more attention to their plaints but also because other determinants of bargaining power such as creditworthiness tend to be correlated with reputation), above by s* i.e., < > ^f- and ^- < 0. Assume also that s* is bounded below by «!_> and 1.^^ Firm Behavior 3.4 Reliable firms and clients always pay for any overrun that they generate. always choose a low level of overrun - yp for the firm and yc follows that they It for the client. At the end of the contracting period they observe the overrun generated by the other party and also whether or not they agree to pay for is it. If a dispute: then the firm the other party refuses to pay for the overrun and the client end up splitting the overrun reliably reliable firms and respects and choose yp and to go to the dispute (i.e., to pay for all of outcome otherwise); and it pay to outcome otherwise; for all of to choose it in the ratio s*/(l They have Unreliable firms want to maximize their short run earnings. mimic the yp or yp and — four options: yp but to act reliably in and all made by contract, as well as unreliable firms and clients will to other to go to the dispute to go directly to the dispute outcome. Unreliable clients also want to maximize short run earnings and face a similar trade-off. actual choice s*). as long as the client behaves as long as the chent behaves reliably to choose cither has generated there it depend on the s that is The written into the on what they expect the other side to do. The following proposition gives conditions under which unreliable firms and chents indeed are unreliable of overrun and then get into a dispute: Claim 1 The unique and yc = yc ""^ equilibriuTn behavior of unreliable firms and i.e., choose a high level clients is to choose yp = yp go the dispute outcome subsequently, as long as the two following conditions hold: Given that \vc have assumed risk ncutrahty, the natural interpretation for s* of the overrun borne by the firm in the event of a dispute. The that it is the expected share actual share will presumably vary according to exact circumstances of the negotiation process, and, on occassion, 16 is may turn out to be 1 or 0. (i)s*{ijF-yF) (a) < Bf and yp and yc 1 ( < Be -s^) (yc -yc) dose are both sufficiently to 0. for there to These conditions arc not necessary be an equihbrium with unrehable behavior but they have the advantage of being easy to interpret. that even if it The were possible to give incentives separately on induced to make their yp- first condition essentially says and yc, neither party could be best choice purely on the basis of those incentives first most extreme contracts in our setting renegotiation rules out the (s* = - simply because = or s* 1). The second condition illustrates an additional source of incentives that arises Ijccause of the specific structure for we assume some part - by choosing to dispute, the disputant necessarily takes on responsibility by the other party. of the overrun generated If the other party was reliable, he could have avoided this part of the overrun by behaving reliably himself. This might give him The second condition a reason to behave reliably. incentives, since it says that if the other party is in effect rules out this particular source of reliable his contribution to overrun is very small. For most of this section wc will assume that both these conditions hold and, consequently, unreliable firms We and clients choose high levels of overrun and then go to the dispute outcome. realize that the first of these conditions later that our results convenient, since it is actually quite stringent. However we on contractual choice do not really turn on this property: limits the number will suggest it is simply of possible cases. Matching 3.5 For most of this section we assume that firms and clients are matched randomly with each other. This of course docs not apply to the clients already worked with Even in the there the question more common case especially since The - - time, may be arc going back to a firm that they have whether the original match was more or of first time matches, however, one as will be pointed out later possibility of selective specific needs is who matching is, may approach with a request random. expect some selection, there are benefits from matching appropriately. - however, limited by the fact that clients often have quite and the number of firms with the capacity to meet those needs quite small. less at any point of Moreover, clients often rely on hearsay in selecting which firms to for proposal and this may imply more 17 or less random selection from the point of view of everything that is pubUcly observable. In such cases, random matching may not be a bad approximation to the truth. Reputation 3.6 Reputation in our pretations of what firm (i.e., have it model is reputation for being reliable. There are at least two possible inter- makes a firm one interpretation reliable. In an it is intrinsic property of the the owner of the firm genuinely prefers to be reliable rather than rich) and and others do some firms not. In a second interpretation, all firms arc greedy, but restrain their greed in favor of better long more pressing immediate some arc patient and therefore able to term outcome and the needs). In other words, reliable firms rest are impatient (or and have clients play a strategy that corresponds to a good equilibrium of a repeated game while the unreliable play strategics that maximize short run profits. Clearly this would only work the history of past behaviors becomes pubhc. been unreliable the future. in with who is known presumption is there is some mechanism by which therefore posit that it if a firm or a client has becomes public information at some point sustained by the threat (say) that no firm will contract been unreliable in the past and vice where the firm and the vcrsa.'^'^ mechanisms by which reputation can client evolve. have contracted at least once before, the that both had behaved reliably so that they both a vis the other. '^^ whom to have is cither interpretation, there are several First, in those cases vis with some probability The good equilibrium ^'"^ a client Under in the past, We if now have a better reputation Forty-one percent of the contracts in our sample involved a client with the firm had worked already. This proportion is roughly the same among young and old firms. Second, the age of the firm should be a source of reputation. We have already assumed that a firm or a client has been unreliable, with some probability this becomes public information if some time at since it is in the future. known Once that happens no one to be unreliable and it will will want to contract with it any more probably end up exiting from the industry.^^ This ^"This might happen because a disgruntled employee reports what really happened or an incriminating docu- ment gets to the wrong hands. •'"This formulation closely follows Tirole (1996). Sec also Kandori (1992). model of how reputation evolves within a specific relationship, see Rauch and Watson (1999). .Ti Under our second interpretation above, the equilibrium strategies actually dictate that no one contracts with "For an explicit 18 more selection process ensures that older firms (and clients) will typically Ijc reinforced by the fact that information in the industry a long ISO 9000 time and does not have any black marks against it, may be certification, are therefore because they think that will it contracts in the sample were done firms (only 9% is more likely to be able to establish a reputation by demonstrating that they follow pro- by outside agencies, such another potential source of reputation. "^^ As we noted the introduction, firms in the industry arc currently very keen to acquire cisely is revealed over time and therefore a firm that has Ijcen cesses which, in principle, should reduce overrun. Process certification as This percent of the firms in our sample were created in 1993 or after. reliable. Fifty-seven Finally, firms is reliable. of the ISO in certification, pre- improve their reputation. Nineteen percent of the external bj' ISO-certified firms. ISO-certified firms tends to be older young firms have ISO certification). Theoretical Results 4 4.1 The Optimal Contract Since we have assumed own that it is the firm proposes the contract, and the firm knows type, the proposed contract can be used as a signalling device unreliable will prefer a contract where to absorb such equilibria clients Among we off will little of the overrun is known to have to be unreliable should certification is expect that there will be than they would be pretending to be the set of pooling equilibria a firm or a client that ^'ISO pays very it less therefore, reliable. is money and equilibria. However all the unreliable firms and reliable. first interpretation firms and clients that therefore should be some approved follow specified procedures to report on the progress of the software more likely to exit. routines. In particular, the firm and to perform the tests. software development process should be easier to monitor for ISO certified firms. Moreover, in a while, and lose the certification if methods. This should give strong incentive to the ISO problems Given that wc awarded by international or Indian agencies, themselves accredited, which examine that the processes of software production in the firm follow§ monitored every once by choosing focus on contractual outcomes where the total joint behaved unreliably. Under the make still we and all its a firm that plans to be - many pooling since in a separating equilibrium will involve would be worse known it most of the overrun a firm may be able to signal that are in a signalling environment, arc who 19 Consequently, the certified firms arc they cannot prove that they followed the approved certified firms to stick to reliably. ISO must standard procedures and report surplus of a firm and a client who Nash equilibrium (sustained by the arc both reliable is maximized. This This expression for joint surplus -dcVF - = V - epyc(1 - - {I er){l is is the surplus that 0c)s*{s, Oc, 9[r){yp + y^). lost is because the firm must allow is because in a pooling unreliable, while the for the possibihty that the client unreliable. To optimal contract we need to maximize the above expression for find the W with respect to This expression or either It is is independent of is decreases in 0/? (1 — 9c){yF +yc)] > and increases in 9c 0. The gives us: always either a fixed price or a time and material contract when most clients are reliable while firms are fact, (s more the reverse situation. reported above, that these arc the only two types of con- also confirms the intuition, given above, that fixed-price contracts are instituted to have a high reputation 'The reader may feel will get is firms. Firms that time and material contracts while the rest of the firms will not. is partly true, in the sense that a less extreme contract will be chosen. effective contract and material contracts protect that the result that the contracts arc always at one or other extreme assumption of risk-neutrality. This is — 9p){yc + Vf) — s, protect clients against opportunism, while time contract - dc){yF + yc)]-gj and a time and material contract in This accords well with the It (1 if [(1 a fixed-price contract likely to be opportunists, tracts.'^'* and only if optimal contract 1). Differentiating ds* - OF){yc + yF) - positive is fact that this last expression Claim 2 The W. s gives us the expression: [(1 is lost is also - s%s,9c,eF)){yc+W) third term in this expression gives the total surplus that term it is an obvious focal outcome. equilibrium a reliable client must allow for the possibility that the firm fifth fact that is: W{s,Bc,9r) The it always a Baycsian- The belief that only opportunists deviate). Pareto o])timal from the point of view of the reliable types makes is However note never very extreme. Therefore it also that s* may consistent with optimal risk-sharing. 20 well if is is driven by the the two parties are sufficiently risk-averse always strictly between zero and one be the case that the initial - the choice of an extreme The Sharing 4.2 The expected Overrun of share of the overrun paid by the average firm with reputation 6p that works for a chent of reputation Oq, is: epOc Vfi hi the previous section an increase effect term on and through and sizes of s* —=— fLxed-price contract or even we showed that the optimal its effect on above expression to the in the if it - OfOc) (1 {s*is,ep,ec)) s is s. An first increase in The term. a decreasing function of Op. Therefore is 6p reduces the share of the overrun paid in s* ^+ yc + for 6p by the firm both through depends on the For firms with a relatively low reputation, the contract . and positive be negative and could s* will it be high. Therefore should be small. in principle For this third effect verj^ is is may relative likely to reputed firms, however, this be a be negative likely to also effect be negative enough to counteract the two other such firms an increase in their reputation direct from the second also shifts weight effect of this shift its might effects: for actually increase their share of the overrun. To summarize: Claim 3 If two firms are the higher reputation relation may matched with is likely to not hold is who have clients the same hear less of the overrun on average. reputation, the firm with The one case where this With random matching, the negative relation for very reputed firms. between reputation and the share of overrun also holds without controlling for the reputation of the client. The mean overrun generated by a firm of reputation 6'FyF This + (l 6p is given by the expression: -dp)yF- clearly decreasing in 6p. Likewise, the overrun generated is in the client's reputation. independent, it With random matching, follows that the total overrun is by a client is decreasing since the reputations of the also decreasing as two parties are a function of the reputation of the firm. Claim 4 The overrun tion. generated by the firm (client) decreases with the firm's (client's) reputa- With random matching, the the firm 's total overrun generated in a relationship reputation. 21 is also decreasing in Extensions of the Model 4.3 Non-random Matching 4.3.1 The assumption there that firms and clients a repeat match. is words compared to the among firms than it The of clearly indefensible for the case the is reasonable to assume that since there are clients"^^, new contract will be random matching is more likely to - where in other to be higher. The therefore potentially many more the firm's reputation will improve by firm's share of the overrun should also go The assumption is contract will differ from the initial contract client's reputation. In this case, the contract. random match between them, both Oq and Of ought may be among at repeat match clearly signals high levels of mutual respect initial new direction in which the amljiguous. However A match start-ups more than the be a time and material down. questionable even for first time matches. Firms with low reputations expect to get a fixed-price contract. Therefore, they have the most to lose from being matched with a client who has a low reputation. Formally this is captured by looking at the properties of the function S = V- 0Fyc-{^ -dcvz - Now if 6c is (1 - ^f)(1 - s*(s(%,0^)),0c,M)(yc + y?) - 0c)s*{s{ec.eF),ec.eF){yF: + m- greater than 6'q \v*{ec,ep) - w%9'c,9f) = - eF){s*{s{9c,eF)),9c,9F) (1 -s*is{9'c,9F)),9'c,9F)){yc -{&c - 0c)yF + {ec + yF) - e'c>*{si9c,9F),ec,dF){yF + yc) +9'cis*{s{9c,9p),9c,9F) - s*isi9'c,9p),9'c,9F)). Ignoring terms that depend on the second derivative of the s*{-) function (the last term in the above expression), the effect of an increase in 9f on W*{9c, 9f) — W*{9'q, 9f) '57% of the firms were created before 1993, while more than half of the 22 client are fortune is unambiguously 500 firms or equivalent. An negative. '^^ unreputed firm (client) benefits than a more reputed firm (firm) While it is more from being matched with a reputed chcnt (client). beyond the scope of paper to explicitly model the matching process that this leads to this outcome, the implication of this result be more likely to be matched with reputable is that firms with low reputations should and vice clients versa.**' It is easy to sec that this kind of matching will reinforce our results in Claim 2 about contractual choice same higher 6p and a lower 6c always go in the - the effects of a direction. Turning next to the average share of the overrun borne by the firm, OpBc Gp^c) {s*{s,6f,0c)) ^vc sec that the fact that j reinforces the result given above in The effect of While case. generates We Claim an increase 6p on 3, so the net effect summarize these may not is is ^^^'s now (1 ~ associated with lower 9c's increasing in 9c- however no longer unambiguous more overrun, Ijc are 1" its partner, the the one given in Claim in this more reputed client, 4. results in 5 Less reputed firms will try to our previous results on the sharing of overrun. since s* total overrun a less reputed firm generates less, will reinforce clients in Claim higher —=T Clients match with more reputed who work with more reputed firms. less As and vice versa. This on contractual choice and the effect of firm reputation who work with clients reputed firms will generate less overrun than a result, total overrun is no longer necessarily decreasing as function of the reputation of the firm. Choice of Projects 4.3.2 The fact that firms with low reputation pay for most of the overrun should clearly influence their choice of projects. This can be introduced into our model by making the plausible cissumption that the most rewarding projects (the ones with the highest possibility of large overruns {yc this large). It is will also have the highest easy to show by introducing assumption into the model of the previous section, that keeping the reputation of the fixed, less "'This s' and Yf are going to be V) reputed firms will be more willing to trade follows from the fact that s'{s{ec,dF)),Oc,OF) off - a lower V for client a lower yc than more s'(s{e'c,dF)),9'c,dF) is positive, while {s{0c,0f),Gc,0f) goes down when Of goes up. Since client wc do not have data on generated overrun is client reputation we cannot smaller for younger firms, which 23 is directly test this prediction. We consistent with this prediction. do observe that Formally this reputable firms. seen by difTcrentiating the expression for ^V*{9c-,6f) with is respect to yc- This yields -{i-ec)s*{s{6c,eF),ec,Bjr), which is clearly increasing in 6^- less reputed firms benefit Here we have assumed that especially implausible: client side Y2K opportunism is is it a is This is not On It is the other hand the firm always has the what hap- interesting nonetheless to look at always accompanied by an equal reduction in yp. Differentiating W*{6c,0p) with -{1 - ep){i jTf- projects are sufficiently simple and standardized that the scope for probably very limited. reduction in yc the expression for reduce yc without affecting possiljle to option of simply doing the job lackadaisically. pens when more from a low yc- respect to yc and yF - s*{si0c,er)),6c,ep)) - yields the expression - ec)s*{siec,Bp),ec.er). {1 Differentiating this expression with respect to 9p, yields (1 - {s{ec,0F)),8c,0F)) s + [Be -Of) :t7, which, for low reputation firms, should be positive (since 9c us that even benefit if — 9f should be yc and yp move together, the youngest firms most from choosing a low yc (and yp) project. Claim 6 Low will This tells probably be among those who positive). '^'^ reputation firms will tend to be specialized in projects which have low potential for client-side opportunism. In terms of what we observe, projects which arc simple work to make clear reputable firms and well-understood so that the is easily defined. may Y2K client who will choose does not have to do very much cither short projects"*^, or projects projects arc typical in this respect.^'' also prefer projects with low fact that these projects protect the clients, The complexity seems to suggest that low reputation firms what he wants. These projects can be where the main goal "Very this yc and yp in this case will of a software project increases sharply with - the benefit in that case comes from the be the vulnerable party. its size, sec Pressman '"Other projects where the objectives arc relatively easily defined include existing software from one platform to another. 24 CAD (1997). projects and migration of an The possibility of switching to a low overrun It yc project should not our model, since yc never enters the expression in does however makes mean overrun will more it likely that the results be ambiguous. Because firms with low reputation may go up with therefore, total overrun reputation. scope for opportunism on the part of the firm, the fall share of the overrun.''^ when will choose projects so as the firm has no reputation and Moreover, if these projects have also less mean overrun generated by the firm not with reputation. Under the assumptions made so tunistic behavior effects to the on the other contracts only protect the contracting parties from oppor- far, side. They have no model by assuming that when be close enough to 1 s is direct incentive effects. that even unreliable firms will prefer to act reliably. client to act reliably rule out the possibility of getting the first-best still We can add incentive chosen to be high enough, the resulting s* symmetrical possibility of inducing an unreliable We In this ease, once s by setting it contract with a lower it further. will still On s. s low enough. outcome by assuming that in order to weak incentives. high enough to induce reliable behavior from unreliable firms, there is be no reason to raise extremes. However will also allow for the give any one side the incentive to be reliable, the other side has to be given very will may Contracts for Incentives 4.3.3 We for the firm's on the relation between reputation and to limit overrun, client generated overrun will be smaller always affect the firm's share of the never behave unreliably, it In other words, the optimal contract be the case that an increase in Op and a may fall in not be at the 6c will favor a the other hand, since the least reputable firms in this world will turns out that for a certain specific range of values of 6p (those exactly around the value at which the contract switches from a high s to a low s) firms with lower 6p will both generate less overrun and bear a smaller share of it, compared to firms with a higher Op-'^^ 'This is not strictly true. the sharing of the overrun. this cfTcct It may A lower yc and yr may affect the choice of the optimal contract and thereby affect However, since most low reputation firms almost always have fixed cost contracts, not be very important. should be noted that this perverse consequence of the incentive effect where the firm (and the client) quite as dramatic and hence it make may discrete choices. With a smoother is most likely in model like set of choices, the incentive effect is ours never never be the case that more reputable firms actually pay a higher share of the overrun. 25 Evidence 5 document that the central imphcations In this section, \vc of the model arc consistent with the data, by showing that contractual forms as well as the actual sharing of the overrun vary with characteristics likely to be correlated with the reputation of the firm. Wc other i^redictions of the model match with the software data. Finally, wc consider some obvious then examine how the alternative explanations of the patterns observed in the data. Measures of Reputation 5.1 Wc have already suggested three alternative measures of or not it We is also in a make use of a fourth metric companies) and external projects. Wc be similar to the difference between there is clearly - wc compare first time and repeated relationships, like in a behavior in this type of relationship after the client has spent a very long is much more the comparisons to firms that perform is, some US together."''' Second, the scope limited, since both parties share in part, a substitute for reputation. OSDCs office of for internal clients arc will be established only the firm. We therefore restrict internal projects (e.g. subsidiaries that works for Note that wc think of the reputation as being an attribute of the firm, more than of the individuals who could be that an experienced professional leaving his job to create a software firm takes his individual reputation with him. what the past career It of the person turns out that individual reputation seems difficult to transport (wc asked who founded the software firm was, and examined whether this was related with sharing of the overrun, but did not find that this was the case). CEO First, time studying the firm. Fully-owned subsidiaries are often run by people who had been previously working in the the two reasons. for know much more about each other (and what they know must be potentially very different from other companies. In particular, It and mother repeat contract but even more strongly, However, we need to be sensitive to the fact that companies working it. whether would expect the difference between external projects to the control rights. In other words, this kind of relationship compose OSDC internal (projects for they have decided to establish a long term relationship for unreliable its age, certification.''"^ an element of reputation: much the firm and the client must I)ositive) if ISO repeat contract and a firm's reputation: of a software firm has to provide is the management The main reason of the team, which is may that the important input or may not be related to his ability as a software professional. The internal/external comparison comparison - since the matching in is therefore closer to a repcatcd/first-time comparison than to a young/old an internal project is like 26 that in a repeat project. their mother company and also for external client). This insures that the selection of firms for internal projects does not invalidate the comparison (since some selected for internal work). Choice of Contract and Sharing of the Overrun 5.2 5.2.1 An firms in this sub-sample have been all Structure of the Contracts implication of the model ex ante rule that contractual forms will be restricted to contracts where the is that firms will bear cither is this implication rests all or nothing of the cost overruns. on the particular assumptions we have made, Ijut it As pointed out, matches well with the observed pattern. As we describe in Section 2, there are three major types of contracts: fixed-price, time and material and "mixed" contracts. Fixed price contracts arc linear contracts with s = 1. As we = 0. In mixed discuss above, time and material contracts arc similar to such contracts, with contracts, the initial agreement specifies a specification phase, another kind of contract sharing rule is eff"ectively splits chosen, which and fixed-price for the In other words, contracts. It agreement is is is payment for the specifications only. specified for the At the end of the development and testing phases. This the projects into two sub-projects. For each of them, a separate either 1 or (often, time and material for the specification phase subsequent work). mixed contracts arc a juxtaposition of easy to understand why, specifications tend to s when fixed-price the project be written on time and material and the is and time and material broken into these two phases, rest of the work tends to be done using a fLxcd-price contract. In the specification phase, the potential for the client to generate an overrun is extremely large. In particular, when the firm first sends the specifications, he can The whole effort important to give the client pretend that the specifications written do not correspond to what he wanted. of the firm until that point becomes higher powered incentive. On it is the other hand, at the time the second sub-contract large part of the uncertainty about (in writing) to the specifications. to the firm, can in effect useless. Therefore, what the client really is is written, a resolved, since he has agreed Therefore a fixed-price contract, which give better incentives become optimal from that point the second phase of the project wants is on. In practice, the choice of the contract for often endogenous: 27 if the firm feels that a substantial amount of uncertainty remains, it can in general insist on getting a second time and material contract. Mixed contracts arc therefore ex ante more constraining for the client than for the firm. Reputation and the Choice of Contract 5.2.2 The reputation share of the overrun likely to both which contract of the firm determines it will end up paying (actual s*). it will get (choice of s) Firms without a reputation and what will be have fixed-price contracts than time and material or mixed contracts. more fixed-price contracts should bear Finally, conditional be more Firms in of the overrun than firms with other types of contracts. on having any particular type of contract, firms without bear more of the overrun than firms with a reputation. The combined a reputation will effect of these that firms without a reputation will bear a larger share of the overrun. is, of course, '''^ This sub-section presents data related to these implications. We presented evidence that age does matter in the introduction, as a motivation for this project. The relationship is and the share of the overrun illustrated in Figures l^orne 1 and 2. and old firms (created in In column firms. external firms. In columns (2) to (4), or after) difference is (1), for 1993 or before). wc report the mean is 26% Young higher). firm. each type of firm, and the difference for the sample of we show the contrast between young firms (created have fixed-price contracts (the probability of the overrun both of fixed-price contracts by the firm are increasing with the foundation date of the Table 5 shows the means of the firm's share of the overrun between low and high reputation The proportion firms arc significantly They more in 1994 likely to also bear substantially more on average (19%), and within the projects with fixed-price contracts (the 13%). The pattern is less clear for fixed-price contracts ISO certification: ISO-certified firms arc not less Hkcly to get and they do not pay for a lower fraction of the overrun in general. However, conditional on doing fixed-price contracts, they bear less of the overrun (20.4%). A ''It relationship with a client has the should be emphasized that the the cfTect of reputation per se. full efFcct power will effects is a general reputation. Firms engaged in a of reputation potentially includes things that arc not necessarily may effect of reputation on the actual sharing of the overrun include a bargaining power effect because the determinants tend to be correlated with the determinants of reputation evidence on the choice of the and other effect as As pointed out above, the conditional on any particular initial contract, of bargaining same initial less likely to like age. In this sense, the contract tends to be "cleaner". This kind of conflation of reputation effects be the case when we arc comparing internal and external contracts. 28 repeated relationship with their client are about as likely as other firms to have fixed contracts, but they pay significantly among Finally, firms less of the overruns who have (20% internal contracts, firms they deal with external clients than when they external contracts are fixed-price contracts (a contracts among They pay a In pay for more of the overrun when deal with internal clients. Almost half of their number close to the proportion of fixed-price of the internal contracts are fLxed-price contracts. smaller share of the overrun (20% instead of 47%) in internal contracts than in external contracts. but this 23% old firms), whereas only much less). The diff'ercnee conditional on doing fixed-price projects is not significant, probably due to the small number of fixed-price contracts among internal projects. is summary, it seems that young firms, firms working with a new client and firms working with an external client bear a larger share of the overrun compared respectively to older firms, and firms working firms engaged in a repeated relationship pany. Wc low, but the if an OSDC or their interpret these results as showing that reputation does influence the are shared between the client projects, for first and the firm. We will way the is that these firms do different tj'pcs of different types of incentives or entail different types of risk. For old firms do mostly project to do the project unless they where there know they is a possibility of very large overruns, they be covered will o\'crruns address some alternative explanations be- possible caveat to this interpretation which require mother com- in case this happens."*^ Table 4 shows that young firms, non ISO-certified firms, and firms working example may refuse In particular, for external clients do on average smaller and simpler projects than old firms, ISO firms and firms working internal clients. It is is for therefore important to check that the simple contrast between the groups not an artifact of the different composition of their contracts. In Table 6, wc show the project-size cells (panel B) differences between the overrun paid and complexity uncontroUed difference of Table 5. cells In panel C, (panel D)."*^ we show The for first a crude way of projects to take into account the and that the two in panel reproduces the the "controlled contrast": this a weighted average of the differences between the young and old firms where the weights are given by the fraction of projects by each type of firm is simply in the project size cells, falling into this project size cell. This is facts that different tjqjcs of firms choose different type differences across young and old firms are not necessarily the same "'Wc will comment more on the choice of project per se below. ''We used the subjective complexity measure given by the firms. 29 for project sizes. all Firms tend to bear simple projects. There of the o-\'crrun paid for for less of is the overrun also a by the when they do complicated weak relationship between the firm. Young old firms is slightly smaller when they do size of the project and the share firms pay a larger share of the overrun than old firms small and large projects, but not for medium-sized projects. young and jnojccts than The controlled contrast Ijctwcen than the simple difference, but high. still The controlled contrast becomes positive, though insignificant, for ISO-certified firms, mainly because the ISO- doing small projects do not pay any of the certified firms does not affect the difTerence between repeated and size Whatever the complexity external contracts. than old firms, firms working with a new repeated ISO and firms client, certification tiear less new of the project, client Controlling for project overrun.'^*' clients and between internal and young firms bear more overrun bear more overrun than firms working with a overrun when they do internal projects. The evidence for once again, mixed. is, In summar}', even after taking into account the size of the projects, firms with low reputation bear more of the overrun than other firms (although the evidence in favor of ISO certification remains A less than overwhelming). evidence final piece of is presented in Tabic different kinds of reputation arc substitutes. still bear more of the overrun we present the (line 1) and difference in the certified firms, if Young some The '"'This Mc A whether young firms benefit from another kind of reputation. In the tabic, in the proportion of fixed-price contracts new who do some Non non ISO-certified old but among ISO-certified firms, there client, more but not if internal contracts, the firms, likely when they work with an ISO-certified young firms bear 27% they have already worked with this same contrast appears: young ISO client. firms are external client and pay for internal client or an caution, as very few is no firms OSDC^^ do small In panel B, Among more Hkely more of the we perform projects. Millan and Woodruff (1999) report comparable evidence in transition economies. 30 jobs than old firms to have fixed-price contracts when they work with an number should be taken with Johnson, in panel contrasts arc interesting. to have fixed-price contracts overrun, but not we examine whether the clients, internal/external contracts (for firms firms are significantly they work with a firms that do Namely, wc ask between young and old firms repeated/new of the overrun than difference. In this table, share of overrun they pay (line 2) within groups of ISO-certified/non ISO- for internal firms). more when they 7. Sellers arc the same exercise, but we look at how the difference between the share of the overrun paid for by firms working with a new rather than repeated cHent varies across different kind of firms. Interestingly, a very different pattern emerges. Tlic difference persists for old firms and for ISO-certified firms, mechanism of reputation formation some for W'^c time, much remains between new and repeated and docs not decline. It clients suggests that the rather inefficient: even after a firm has been in the market is to be learned about it. have documented systematic differences in the way cost overruns are shared across young and old with repeated and new firms, contracts and firms reputation in internal is clients, ISO-certified firms and external contracts. This evidence is and other consistent with a firms, model where an important determinant of the contracts and the sharing of the overrun. In the next subsection, we examine the whether the other empirical predictions of the model also hold. Further Results 5.3 Choice of Project 5.3.1 A simple extension to our model also predicts that the firms with a low reputation will tend to choose simpler projects where the objectives are We the overrun generated by the client. Table 4, and Figures 5, 6 and 7. Young easier to define, which will tend to limit present evidence relating to the choice of project in firms do smaller projects (Figure 6), which have smaller overrun (even expressed in proportion of predicted costs). ^" They also tend to carry more often "simple" projects (Y2K, arc easily defined, CAD, and are data manipulation) (Figure We easier to monitor. 7), which generate lower overruns, have also asked them to subjectively rate the complexity of the project, and even according to this subjective measure, young firms do more simple projects (Figure project (cost multiplied by This could be at 5). As a markup) is result of these two combined facts, the returns from each smaller on average for young firms than old firms. least partly explained by the fact that young firms arc on average less com- petent (and that therefore clients do not want to entrust them with large or complex projects). more willing to extend trade credit to a or business network than if now customer if they have obtained information about him from a social they have not, but the weight given to this information declines as the length of the relationship with this buyer increases. ' Moreover, by doing that they keep the share of overrun accounted for by each project more or across young and old firms: this could therefore be explained by adding 31 risk aversion to the model. less similar However the same contrast holds between and size Y2K projects. Since internal work, the difference this bias. we have internal and external projects (Tabic restricted the 4), for comparison to firms that do at between internal projects and external projects This confirms that part of this difference between young and old is project least some not tainted is Ijy due to difference in behavior. Overrun generated by the firm and by the cHent 5.3.2 Our Ijaseline model predicts that, on average, firms with a Ijetter reputation will generate o\-crrun (and therefore, total overrun will also tend to be smaller for their project). less However, once wc take into account non-random matching and the choice of projects, the prediction about the relationship between overrun generated by the firm and reputation arc not unambiguous. On the other hand, these extensions predict that clients matched with firms with low reputation should generate less overrun. Table 8 presents evidence on overruns generated Ijy the firm and the client. ^^ There is systematic relationship between total overrun and the reputation of the firm: total overrun smaller for \'oung firms, firms without firms dealing with a new generated by the firm but larger is client. argument that less is certification, of these contrasts slightly larger for for ISO-certified firms generated by the client None ISO and external is all it is close to being significant. client. ^^ cases for less reputed firms, wliich is is larger for Overrun young firms and firms dealing with a new and firms dealing with an internal smaller in projects; no client, Finally, overrun consistent with our reputed firms try to protect themselves from absorbing large overruns by dealing with reliable clients or by choosing projects where the scope for client generated overrun is smaller. The data seems consistent with our model of how reputation determines contractual out- comes. In the next section, we examine the most obvious alternative explanation to the observed pattern. Overrun generated by the firm and overrun generated by the among do not necessarily add up to total overrun: sources of overruns proposed to the firms, there was an 'other' category, which ''"These results could be consistent with the idea that overrun First, there as client is in general no compelling evidence that wc have noticed ISO is slightly lower when we left unattributcd. firms have a good reputation. certification really gives a firm a reputation. Second, already, overruns are less costly for internal projects, so they tend to be larger. 32 Alternative Interpretations of the Data 6 This section reviews alternative explanations to the pattern young firms bear a to the main 6.1 Pure Risk Sharing One possible interpretation of result that as explained in much what going on in this industry is showed // the (in particular is pure risk sharing. However, greater detail in a previous version of this paper, this interpretation very where there is we assume CRRA preferences (which is standard we substantial variation in the size of the contracting parties) in the previous version of this Claim 7 data larger share of the overrun than old firms). quickly runs into trouble. In the case where in cases like this ol:)Scrvcd in the paper that firm and the client have CRRA preferences, for a fixed project size, the share of the risk that they each bear will be approximately in the inverse proportion of their coefficients of relative risk- aversion, keeping fixed the ratio of their total revenues. It will also be approximately in the direct proportion of their total revenues, keeping fixed the ratio of their risk- aversions. Given that the this proposition risk averse is client's revenues are much bigger than that of the firm,^'' an implication of that the client should bear most of the risk unless the client than the firm. In fact, is much more the firm bears on average more than half of the cost overrun, suggesting that the client's coefficient of risk-aversion must be very large relative to the firm's. is however difficult to It think of a basis for such differences in risk aversion. Moreover, this result has systematic predictions about the relationship between firm size and the share of the risk that it bears, controlling for client size paid by the firm by client size, in size. Tabic 9 presents the share of overrun project size and firm age. In firms bear less of the overrun than shown and project young firms. Figure 4 and in columns 2 to 4 in Table all 4: old firm's turnover is 26% is larger are with $ 10 million. Among old Small 27% is million, sized client, median turnover of the software companies only $1.2 million, and the largest firm had a turnover of $47 million. Only above by $3.7 medium arc with small clients. Large clients are in general fortune 500 companies or equivalent. firms with turnover below $10 million. In contrast, the cells, Since old firms are on average larger (this ''^Morc than half of the contracts in the sample arc with "large" clients, 19% project size-client size and clients are in the sample of the firms have a turnover the firms engaged in contracts with small clients, the median firm has a turnover of $0.5 million. 33 or inoic than 100%, than young firms), imphcation of the risk-sharing this contradicts the basic model. One miglit also speculate that old explains why and new firms generate different risk profiles and that old firms bear less risk: perhaps old firms simply generate less risk. However, the evidence on the standard deviation of total overrun presented in Table 8 shows that this the ease. firms. more The standard There is is deviations of total project overrun are very similar across therefore no evidence to support the view that Another risky to deal with. of o\'errun this different for are not different), possibility, and that the particular form of the We young and old firms (Figure examined the 8). The two not types of young firms are systematically however, would be that the underlying distribution young and old firms (despite the risk sharing rule optimal. all is fact that risk faced mean and standard by old firms made de\'iation this particular entire distribution of overrun generated for both distributions arc very similar, except for four old firms which generated very large overrun (150% and higher). These four old firms are however not driving the results, since all of them paid 100% of the overrun. Moreover, above, the difference between old and young firms type (complexity or The evidence we of course, many size), is maintained when we control which arc presumably good indicators of project-specific give above strictly only applies to the case of classes of risk preferences which do not fall on the one very hard to explain falls slower than a CRRA why the firm bears any risk at (so that the preferences that project size and client size effects also all. risk. There arc, into this category. However, there work particularly than a side, if the coefficient of absolute risk aversion falls faster it is for project CRRA preferences. are two basic intuitions which suggest that these other preferences will not well either: we have shown as On approach the become smaller and the other side, CARA model), CRRA, if the coefficient it can be shown this leaves very little to explain the inter-firm differences.^'* 6.2 Vcirying Levels of Competence Our model has assumed that both that firms and clients contracted upon. '' clients and firms arc make mistakes which In this case, one possible In the extreme case of CARA risk-neutral. Suppose we now assume lead to overrun but that these mistakes can be first best contract is one in which firms take the preferences neither project size nor chent size affects risk-sharing. 34 responsibility of full competent, then The it is to be expected that they point against this explanation first assumption. If the firm against the firm was at all if young firms^'' arc would pay on average is for more simply that risk-neutrality risk-averse then the optimal contract sources of risk that are beyond all Now any mistake that they make.^^ its on average less of the overrun. is a very extreme would try control. Therefore, since to insure young firms do not choose to be incompetent, they should be insured against overrun that results from their mistakes. Of to bear risk, but as than the this may course, the extent of such insurance firm, and we have already argued, the in particular small firms client well be limited by the client's willingness is in a much better position to Ijear risk should only bear a small part of the assumes that the mistakes are not made deliberately. The case of varying hazard will tantly), firms in pay much more of the overrun than the share Table average 51% entirely by of 3, it. even when the client is difi'crcnccs in the share of is, if tliis for view: levels of moral overrun that is caused by the firm, more overrun than old firms The in Table to old firms 8. is The difference it is would have to be the between the share of clearly not large more enough Differences in sources of the between the fraction of overrun due to young firms only 1.18%, but they pay 20% more of the overruns. Finally, even within firms that do some internal work (and are therefore more homogeneous), that firms bear pay on do. Recall that the total difference to explain the difference in the sharing of these additional costs. and that due (and most impor- fully responsible for the overrun, the firms still anything, larger for old firms than young firms. overrun are shown first which they are responsible: as overrun due to young firms and the share of overrun due to old firms it course, Second, to explain the differences in the share of overrun paid by the firms case that young firms cause substantially overrun Of be examined below. There are also some simple empirical arguments against shown risk. it is the case of the overrun in external contracts than in internal contracts. Therefore does not seem to be the case that the differences in the share of the overrun borne by young firms can be explained by systematic differences in competence between '' ' young and old Of course the actual contracts do not say anything about dividing the overrun. Therefore what we arc to here ''''Or is a fully efficient implicit contract. more generally, firms that wc have called so far "low reputation" firms. 35 firms. referring Underbidding by Young Firms 6.2.1 One could imagine that even in a world where contracts are efTectively complete, young firms might systematically underbid (quote a price based on intentionally low estimates) to win the Of course the project. them responsible end up paying knows this, and in the we optimal contract corrects the extra overrun resulting from the underbidding.^^ for a higher share of the overrun. objections that 6.2.2 for client list However it for Young it by holding firms therefore should be easy to see that the same above to the competence-based explanation also apply in this case. Varying Level of Moral Hazard The arguments against the two previous alternatives are based on the assumption that the mistakes (or prediction errors) are deliberate. Once we allow for such moral hazard, the client may Old firms could be well not be willing to insure the firm. young firms, and therefore bear less risk than young firms. on mean and variance of the overrun cannot respond to endogenous: young firms could be generating the same less prone to moral hazard than The evidence we have presented this kind of criticism, since they arc level of overruns as old firms, precisely because they face higher punishments. Note however that this would be view only if the levels of moral hazard were about the type of the firm or the reputation model). Moreover, difference that a firm to change very is client it is - if common knowledge there was learning, the if first client. The would just be a variant of our level of and the second contracts that firms arc treated differently the second time around our there was no learning not clear why, in this alternative scenario, working with a repeated much between it (i.e., a real alternative to for it should moral hazard a specific make any is client. unlikely The fact must therefore indicate that there is learning going on about the characteristics of the firm. 6.2.3 The in Dynamic Moral Hazsird fact that repeat contracts are different from first time contracts could perhaps be explained terms of the evolution of a dynamic incentive contract, in the absence of any learning. However this would not explain the contrast between the first-time contracts faced by young and old Moreover "Again it \vc is not easy to see why, in this view, the age effect would be smaller arc referring here to a fully efficient implicit contract. 36 when firms. it is a repeated contract. 6.2.4 Young Varying Levels of Honesty They firms could also differ from old firms in their propensity to report costs honestly. could be more prone to try to report inflated costs, or to pretend that changes due to their incompetence are due to the client apart, then the analysis of such a changing his mind. If model would be similar the client could not to the analysis of the and lead to the same conclusion (the reputation of old firms would be instead of a reputation for reliability). model a reputation for reliability, different reason for the Note however that but As wc mentioned it is earlier, tell own the cheaters model we propose, a reputation for honesty our modeling choice was to clear that the analysis could be carried out with a importance of reputation. if and honest clients could tell apart cheaters firms, and punish cheaters by imposing them to pay more of the overrun, then we would also observe that young firms would pay on average more of the overrun (but this would not result in any social cost, unlike in our model or a version of the model with a reputation for honesty). Assuming that firms report in the questionnaire what they have reported to the client, then the evidence that pay more often than old firms overruns reportedly caused by the fact that they are lying more often than old firms. Because that firms are lying in what they report to us as well as in is client this young firms would simply argument what they say rests reflect the on the fact to their client, it not easily verified or invalidated in the data. Note however that this argument implies that the clients never self-defeating: make any mistake why would in telling apart cheaters firms cheat in the found out? Moreover three facts are first place if difficult to reconcile and honest firms. It is therefore they know that they are going to be with this explanation: First, firms pay on average 50% of the overrun when they report that the client is fully responsible for suggested explanation would therefore imply an implausibly high fraction of cheaters it. The among Indian software companies (young and old). Moreover, if the client has perfect information and can enforce any sharing ex post, there should be no variation in the contractual form, or at least outcome. it should not be related to the final However, firms pay more of the overrun when they have fixed-price contracts than when they have time and material contracts. Furthermore, price contracts. Therefore the ex ante contracts young firms have more often fixed- seem both to be relevant and to be used by the 37 clients, which is not consistent with the world Finally, note that such a repeated and new we just descriljed. model would not explain the difTcrcnce between contracts with and external contracts: chents, or the difference between internal has perfect information, then not easy to explain it is why if the client firms would behave differently when dealing with different types of clients. Conclusion 7 We set out in this paper to look We mining contractual outcomes. though given that the evidence when deciding on tion for is contracts) evidence that reputation plays an important role in deter- indirect (we a this view, do not actually observe people looking at reputa- and there are important firm characteristics that arc potentially correlated with our measures of reputation, The conclusion seems to strongly support find that the evidence that reputation matters some doubts is clearly remain. of course important in itself: range of theories that arc based on limitations of contracting. Moreover, explanation of why the Indian software industry is not much differently, and the why is it fact that this is an equilibrium it gives support to might suggest an larger (Indian software exports were only worth 3.4% of the 1995 worldwide outsourcing business) given advantage'^*' it its obvious labor-cost a very labor-intensive industry. Or, to state the same point for software professionals in India to get paid so than their U.S. counterparts? Reputation at the firm level is much less one possible explanation: most Indian firms are simply not trusted enough to be given important contracts. While our evidence cannot directly substantiate this view, the fact that reputation industry suggests that it also ought to is important within the Indian be important when an American client is deciding whether to go to a firm in India or to one in the U.S. To add support is to this view, our results also suggest that the process of reputation formation rather inefficient. This firm is is reflected in the fact that after controlling for age, dealing with a repeat buyer words, repeat buyers clearly still makes a substantial know much more about whether or not a difference to the contract. In other the firm than the market does. In other words, the fact that a firm performed well in the past vis a vis one firm takes time to become '''The U.S. of imports a very large number of Indian software professionals more than twice what they would earn in India. 38 for short-term assignments at a cost public information. This but it is of course consistent with rational behavior on the part of the client clearly hurts the firm. The policy implication of this view rating systems may aggregate that a credible system for rating firms modeled on credit play an important role in the evolution of industries such as the software industry where contracting efficiently is all that is is inherently problematic, by known about each making it possible for the market to firm. References [1] [2] Bajari. Patrick Plus", MIMEO, Baker C, firm" [3] J. Murphy Working Paper #6177. and John Umbeck (1984) "The effect of different contractual arrangements: the Law and Economics 27:313-328. Crocker K. and K. Reynolds (1993) "The efficiency of incomplete contracts: an empirical analysis of air force engine procurement" [5] Cost (1997) "Implicit contracts and the theory of the case of retail gasoline market" Journal of [4] vs. Stanford University. R. Gibbons and K. NBER Barron and Steven Tadclis (1999), "Procurement Contracts: Fixed Price Gcnesovc D. (1993) "Adverse selection RAND Journal of Economics 24-1:127-146. in the wholesale used car market" Journal of Political Econonuj 101-4:644-665. [6] Greif A. (1994) "Cultural behefs and the organization of society: reflection on collectivist and individuahst A historical and theoretical societies" Journal of Political Economy 102-5:912- 950. [7] Hecks R. (1996) India's Software Industry. State Policy, Liberalisation and Industrial Development, Sage Publication, New Delhi, [8] Holmstrom, Bengt (1982) "Moral Hazard [9] Johnson, S., J. Thousand Oaks, London. in Teams" Bell Journal of Economics 13-2:324-40. McMillan and C. Woodruff, (1999) "Contract Enforcement IvlIMEO, MIT. 39 in Transition", [10] Joskow, (1987) "Contract duration and relationship specific investment: empirical evi- P., dence from coal markets" American Economic Review 77-1:168-185. [11] Kandori, M. (1992) "Social norms and community enforcement" Review of Economics Studies 59:63-80. [12] Lafontaine F. and K. panel data" [13] Mastcn S. NBER Shaw Working Paper #5585. and K. Crocker (1985) "Efficient adaptation provisions for natural gas" [14] (1996) "The dynamics of franchise contracting: evidence from in long-term contracts: Take-or-pay American Economic Review 75-5:1083-1093. Monteverde K. and D.J. Teece (1982) "Supplier switching costs and vertical integration in the automobile industry" Bell Journal of Economics 13:206-213. [15] NASSCOM [16] Pittman R. (1991) (1997) Indian Software Directory, Hill International editions, A Of California at [20] STPI [21] Tirolc, RAND 34:565-589. Practitioner's Approach, 4d edition, in an Unfamihar Environment", MIMEO, San Diego. Shepard, A. (1993) "Contractual form, retailing" rail- Software Engineering Series. Rauch, James and Joel Watson, "Starting Small University [19] Law and Economics Pressman, Roger S.(1997) Software Engineering, MacGraw [18] and opportunism: the evolution of "Specific investment, contracts road sidetrack agreements" Journal of [17] CD-ROM. retail gasoHnc, and asset characteristic in gasoUnc Journal of Economics 24-1:58-77. (1997) Software Technology Parks of India, MIMEO, Bangalore. Jean (1996) "A Theory of Collective Reputations (with Applications to the Persis- tence of Corruption and to Firm Quality)" Review of Economic Studies 63.1:1-22. 40 Figure 1 Proportion of fixed cost contracts 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1995 1996 1997 or before Foundation date of the firm Figure 2 Sfiare of overrun paid for bytliefirm 100 1988 1989 1990 1991 1992 1993 or before Foundation date of the firm 1994 Figure Mean (percentage of 1988 1989 1990 1991 3: overrun of project initial 1992 evaluation) 1993 1994 1995 1996 1997 1995 1996 1997 or before Foundation date ot the firm Figure 4 Average Firm Turnover 1988 1989 1990 1991 1992 in 1997/98 (Millions $US) 1993 or before Foundation date of the tirm 1994 Figure 5: Subjective complexity measure 1988 1989 1990 1993 1992 1991 1994 1995 1996 1997 1995 1996 1997 1995 1996 1997 or before Foundation date Figure Size of 1988 1989 1990 tlie 6: project (In man-months) 1992 1993 Foundation date 1991 or before 1994 Figure?: Proportion of "simple" project (cad, y2k, 1988 or before 1989 1990 1991 web pages, data manipulation) 1992 Foundation date 1993 1994 Figure 8: Cumulative distribution of overrun A: J Young firms 50 B: Old firms O O o E 3 P X — o o o o o o o o o o O O d S o o o o o E o o o 3 ^^ E 'S II Z %r: -> o £" 0. i — O ^ ^ "" ^ ^ so ^ac -n »- O -*. — o o o — in u-i vC r~ ^ — d d 1/^ <N — (N — C> ^. o o wi ^ "* M^. O ^S a c u. O c n '>< UJ r^ M^ o o o o o '5 o S r-* t^ C « o w-j »o r-i c-i ^ P — — — d r- r4 (N m d o t^ oi o) in -^ 00 c\ <Ni ov in M- c> — — — o o o o o o o o c o o d — — oo — vp fN m d d d s O O E 3 o o E ^ « 2 o o o o o E ^^ vC f*^ 3 E 'c II !i o a ex a z E C/l o o i- q rS in "* -o5 T]- d — — d d ^ in m _ m Mo o o \0 ^ ^ d vn -* u-i —C3 O on -a o o o o o o o o E c c oo r^ mi vD 1.2> c. U G\ M- TT oo •3 S oo rS 00 <^ c « a y H * r^ d (N (N vO d o d * •* d (N (N _ od tT OO r~ <N O) r-^ cN o\ o\ d m (S d 1 T — 00 <^ •^ o o 'S- r-* S u iu < < uj O o u ,o ,o « £^ < ^ E w o u A o y —rtn os «CJ C 6 i- ~ y c o o <-o .H, •c t) ^ C- t <D o t;; .1^ M XI E"^ .Ei:^ < o' £ a. S^ 00 S^ a. a. a. o x> .£? i:r O. -o M 3 3 o 3 T3 -O o o o o u u u _ -a _ _ ^ U ^ ^ ^ ^ ^ O a. H m o Zj ^ u m J UJ z < a. H U < u 2 c o u "5 >» ^^ c O a u o D. E o D. D. u « C 1) U ¥ o E Q 00 a o w >< z o u u -J w z < CU c E C B U E T3 a 2 c o u o X o 2 c o o o u •o o X c 1) > o o E a s: 00 XI Table 2 Share of overrun for paid by the firm as a function Proportion paying 100% A: and 100 Proportion paying % (2) (1) PANEL Proporlion paying between of initial contract Average share of the overrun 0% paid for by the firm (3) (4) ALL CONTRACTS All contracts 39.29 17.85 42.86 47.4 (3.58) n.i 11.12 77.78 contracts 34.29 25.71 40 Fixed cost contracts 54.12 18.82 27.06 Time and material contracts Mixed 15.6 (4.92) 51.6 (7.69) 63.1 (4.83) PANEL B: EXTERNAL CONTRACTS All contracts 47.58 20.97 31.45 57.1 (4.06) Time and material 17.65 23.53 58.82 contracts Mixed contracts 28.2 (9.56) 40.62 25 34.38 50.9 (8.08) Fixed cost contracts 56.76 18.92 24.32 65.8 (5.07) r~~~ 1— E E c o CJ u 5 ^ ZJ ^ C/D J o >> x: ;o o. -^ 'J~' CN .^ ^ CM — r^i I— (N ;.- ?3 ^ 2 u^ in CN r~ \o U-] 30" <^. o o ^ 5 oc -""^ -^J O O d O ° 5 o • :::r" '^ t> •3 C o 5 ZJ i) b § ^ c/; c>: O as 00 o ^ t^ c^, Cm r^ 00 in t/; E L. iE p Q C. c/: k. 3 t» k. o 5 ^-^ t) k. E k« T3 T^ /-^ cs 00 >n o "^ >. j= * oo" x: r-^ 1-^ — '^ r~ 0> VC vC CN • • \o ^^ 00 5 o v. C o oCX w: r^ 3 k. CJ > O JJ c O u. >, U S& •5 1— 1 — o b ^ c/: iJ £ '^ ^ ' ^"^ -* ,x CN ^-^ sC CN uo 1/^ ^ o r^ \o :::r^ o vn • CN c c o u c. O k« • m 2 • vd -^ __, c/; 00 U H a Z o u o o h c o u c o u o u u E 1) T3 c u CO H X m i3 CO c o o < D, on 1 o u g X E c o o s f" •3 o ;:^ o c o '-> 0\ oo 8 ^ OJ ^ \0 "" • /-) p „ -^ Q ~ Cl c '_» ':j 2 "^ o o o o S ^ o O o o '-) "^ C: -. 12 — — — c CJ CL ;^ H c M — '^ m ^ O O in o d «y^ "a . tN C.1 2^ -?! o Tf CNi X UJ o rM u r^ — c: d c o o o_ 9 P o d C o So" ,o ?i ' '-» o ^ ^ c. ~ c c b f<-J "D oo o ^ — O o d NO ^ O — d O^ ivn T5 CJ O o d a: c o CJ 0^ o > -^ 4 o o o o o d d £ o _• ^ o — oo o O o _ O d r^i 1/-1 in u 2: o u c u 0) q- CD tn a ^ D _ _ ON o "^ r- — ON (N| ''^ en O •-J f > m m o o o 3^ O » °o ^- ^ *— oo --- -/S t/: ^ ::; oo VO ^ in u. 0) c: o CO ri Q. XI c (-4 — ^ 2 oo -^ in ^ -^ ON _; _j *-' 00 _ in o o O o d O _ o > c/: rN ^ ^ C ^ - p s p S d Oo * (N V~\ r-' r~~ r^i "^ '— r~- CJ t^ m o o d d 4 =^ o o d o o d O oo— pc^ — C — ~'-r — •* o m _• — o — m d o c oo . o o d _ o^ d o o (N o o o 2 m 'O ^i § _ o o ^ in 00 in C <u ^ NO "T NO 9§ ON /-> o "O 5 — o in — — d ^' — ^^ o — "' c _o oo CT (N r^ O CN c 3 £ o CnI dS o o O ^ o d d p s :2; II V E u. n-t 3^ ON _ NO NO m o o d S d o o o d o o — o 1-1 A -^ '^ i B ON I^ ON O d o ^ v> •^ C O >N = >- O — ^^ £ ^ "" C >> _0 5 ^ On tN 2 — — 2 — Tf « O o O o O O ^ d d in S' r^ ^. p o o o O mo ^ — ^^ o o r-. "J- ?7 o.-a Q. »5 (3 o o It E ° c o u £ '-^ • O. Q. o o O c/5 o H o. o •^ E 0. ^ CT\ -p p S o '^ •n 0° Cn) fe. .Si 13 c^ ,29 E (i: o ~ — o <-- o '-) c ° d ;-» .'-t Q 'J v; o C ^ O o O d ~— 00 u~i ~ c. o o u c Ch >^ H E oo ^ od f»i O ^ '^. o m "^ E o X UJ c o c ^ '-) .'_) g S 9 K •-J '~ _) CJ Q _c -rj O o d '~j 5 o CI "cz ~ ~ C. o c o ce: g 5 d s d d c o o a: 'J 1^ Co f> DO s o z — o o c cr _o c3 _ CJ — d CO cs b '~i in t^ §1 -u c --^ ° d o O c« o >- O O O — o d o z Si e: - 2 oci O b c o f^ c> -3 o- ( O o d 3 o V II g oo o tn o ^ d ^ u. . — . — - „ oo VO A 00 c^ 00 ti~t "^ £ ro m o ^ o S d • ": IN ^ e t'l E S u o ^ c o o o n c o CL o > c o o ~ X3 o O o o o <*- rt •O J= '« rt -C "5 t/2 IZ) c o 3 c ^ ^5 . ,-v rj^ ^— . ^ \0 '^ 'P.'^^ <r) CM CO p- r^, oc • o\ o) \0 r-1 OO r^j "'^ vo t-~ ;— rf -- T3- ^ r— '^' — • w O U • C-4 — r-~. c! 2 ^ Co ;f2 s p^ >~~ !2 * _ - tN <N O o i_ — u r-\ — \^ '—- ' "* 00 vo 5S o t~~- ^ d ?i ;^ ^- r^ ^ — ^ ??, :: oc- ^" ' (D N ~o (/) •a u ^ O C. r- c3 0) "2 .£ Q. >. — r^ ^ o ^ ^ 5 ^ — '^' •:-> 'V v: =c gr tG> \o f^ vo vo Q ^^ w o TJ u !*• 0) ^ so :^ r^ 'Ti~~ ^ - -* 5^ ^ X oc r^ '^ r^ .— ^^ — — — '^ oc ^"^^ ; ^ \d <£> 0) £1 CO I- ^ 2: 1- O '^ CJ u c c o l_ a CO 1 ^^ ^'-V 1/-X S ^ 3 o ^ i£ S <N oc —' /'"v r*^ ' o ^"— <^^ 1 I -:r^ -—^ 1/-J ^^ OC •-^ u(— cu 5 u^ CJ o o CI, o o c v; o >- '""' o\ s T3 C g o c; s- ^. c^ • ^. ;:: -; "^ •* "^ ~-' ^ * r^ ^ ^ 00 pq (^ rj- I-~ ^ r~ r^i r~ _; u-i * ^ pi "^ !£2 ^^ c^ t^ o "* '^ .^ so ''^ — ^^ OS jC£ Q m • o\ in • OO cs) ^ — ^ m ^ 00 "J OS so OO "^ OS00 ^ • o OS^ OS >o so Ti" V E E p :i "^ ^ m — OO _; — 2 o u c o k. •o c 3 12 lO _: ?3 '^ 2; VO Z O M S ^ O ^ o c/^ CO — .^ o o ZJ D. C S > o o OC CN r<i OS ON — Sr S ^ PT P 2 ^ S ° 2 -;. 00 vo <^. so c <N '„ ro) OS ^' OO m • o A G. j: ^1e o g E E m J m ^ so a. V E OS 't SO o c o hi c C o E c E o <a- A U U J m 2 < a. o bfl CI. >> D3 Q J > o Oi iS" 3 o QJ m II H^ 3 w ^ *x ^ "<U ^-. m ^ *>< 4> ULl < CLi 1t O U e o 1 o u < 2 r3 > './; Co CJ 'c" jr o t— s "rt S E t- E c ^ o c. o c cc. o * o o c C- H o ?r o ^ ^. 7.9 3.6) ;2 — d d ^.^ >< c oO o "3 o c o — Di 5 c. _ .E y; C "o _a > o o _ ON oo :: o< ^ s ° d - C/3 z ON ° d Q. c H Z UJ J u Q tu < t^ -^ cu UJ Di UJ CD Q Z < c Z H • J3 o r^ Ov tc I— B oo o 11 V s Tj- 2 d H U < H r:" Z o u z w m H O — c ,o oo erf m ON t: ON c 3 o A a, UJ pa UJ u z o u -a u X UJ DC S o c o „ CO ^ H o 2 Q. w n O ^ o On O C/D D. e us m J w z < a, O c/2 XI a u c u P^ °? o -i ^^ -r22 ^iP oo^ =r^ [11,=^ ~ C "V.^ '"' o- o d -r^ ^^ 1— „ CJ > "J^ Q o u _ C- c c o o c —" c. >^ H Tz c d 3 d ON in ^ in o o d o X UJ 00 o — Cr OS •_) H o 'J c: (D II ,o '~ cj O jZ rj O C O o II o o zj > c_ ~ c r; O Qi _o JZ c Qi Tj 01 o o ? o z: n o o "T ~i '_> c o II c (0 >. m or a CJ J. '— O d c o d 3^, *S 0) > O o _ o o sq r*^ O o d Q [— o o c^ O on •^ ^ ~~ II .ii o o rJ oo 5^ ol o o d u: CJ > d o Z o. o — O u^ — o U. C c o II .CJ CJ 1— -:3 Q c o rt oc s ^ vO ^ s \j II m o ov D o V II w. E u. dS r*~, O d CT^ A o o c E 3 o c 'wi D t § > O o "rt 'c £ M o C o 5 E c CJ to ra ^>. £&0 3 E c U Ml 3 n CJ c c 3 u > i E CJ "n 'E •d u5 JD TD 3 ,0 •0 o o o d c: T3 c: >% Xi •0 CJ > CJ T3 i-n 3 ra 3 t ^ C3 c OJ 73 CJ ^ 'o' CL Table 9 Share of overrun paid by the firm, by project and client size Youns Old firms firms Size of project Size of project All clients Small or medium clients Big clients All <median >mcdian All <median >median (1) (2) (3) (4) (5) (6) 68.7 72.5 63.3 46.6 54.7 45.7 (5.15) (8.17) (9.82) (6.25) (8.34) (6.60) 77.5 79.9 73.2 65.4 61.1 68.2 (7.93) (10.0) (13.7) (7.5) (12.7) (9.5) 59.8 63.5 55.8 37.6 50.2 29.3 (13.9) (6.56) (11.2) (7.72) (9.49) (13.5) DEC <^nop Date Due Lib-26-67 MIT LIBRARIES 3 9080 01917 5865