On the Supply Side of the Capital Goods Market: Roles that Maintenance and Repair Can Play Junkan Li The Ohio State University, Department of Economics li.6118@buckeyemail.osu.edu A Simple Model with Analytical Solutions September 8, 2019 1 Introduction Below is a simple partial equilibrium model, which, for now, only contains two types of agents: Consumption Goods Producers (called CP hereafter) and Investment (Capital Goods) Producers (called IP hereafter). The simple model is of finite horizon, 3 periods of time specifically. All the decisions are about buying and selling capital goods and maintenance goods. The simple model keeps most of key features in the expanded version of model: a) Substitution between capital goods and maintenance goods for both IP’s and CP’s; b) A forward-looking capital pricing formula where the existence of maintenance matters; c) An endogenous amplification mechanism due to the possible default of IP’s. This simple toy model is presented to provide insights for understanding the significance of maintenance and repair (M&R hereafter) in modern macroeconomics, rather than to numerically match the data in the real world. Although simple, the model results indicate that introducing M&R has potentials in explaining or resolving the following macroeconomic questions: a) The high volatility of investment and investment-like durables; b) pro-cyclicity of M&R expenditure and maintenance/investment ratio; c) pro-cyclicity of capital good prices in micro level and counter-cyclicity of relative price of capital in macro level. Overall, this model is offering us some inspiring ideas for answering one of the big questions in macroeconomics: what are considered as important on the supply side of capital goods market and how shall we model these? 2 Model This is a finite-horizon economy, with two types of agents: CP’s and IP’s. The measure of each type of agents is normalized to be 1. CP’s are homogeneous and IP’s are heterogeneous. There are three goods in the market 1) consumption goods produced by CP’s by using capital goods only; 2) capital goods, produced by IP’s and sold in the fore-market of capital goods market; and 3) maintenance goods, produced by the IP’s and sold in the after-market of capital goods market. The consumption goods are treated as numeraire. This simple paper is focusing on modeling relations between capital goods and maintenance goods. Following the conventional assumption in the micro literature, I assume CP’s are price takers in both fore-market and after-market; IP’s are, on the other hand, price makers. 1 Compare to most of previous literatures, what is new in this paper is the way I model the M&R, which further determines the transitions of capital goods that CP’s hold. Therefore, I shall start with M&R. 2.1 Maintenance and Repair Each period, consumption goods producers utilize capital goods for production, and by the end of each period some parts get broken. The maintenance & repairing technology here after is modeled as replacement the broken parts of old capital goods with the new parts sold by the original capital supplier. An old capital good can be used for production in next period only if all the broken parts has been replaced by new parts and this paper does not distinguish old but functional capital from new capital. Therefore, given the marginal cost of buying a piece of new capital p, repairing cost cm , and the price of new parts pm , the consumption goods producer would like to repair the old capital goods only if the repairing cost is lower than cost of purchasing new capital; mathematically, if maintenance & repairing is the better choice, then the number of non-operative parts x should satisfies: x ≤ m∗ (pd , pm ) ≡ pd − cm , with cm > 0, pm > 0, and pd > cm pm (1) Another assumption in this paper is that CP’s are allowed to trade-in capital goods to their producers. For the capital they do not want to repair, they can sell them to IP’s and IP’s are capable to find the operative parts and utilize them. Both of CP’s and IP’s have incentive to make that deal. However, since IP’s are the price makers, they will for sure set the price to the minimum level to which CP’s won’t reject; that is zero. Therefore, the M&R behaviors of CP’s can be summerized in the table below: Dealing Behavior M&R Trade-in with an Old Capital Good Condition Benefit ∗ x ≤ m (pd , pm ) pd x > m∗ (pd , pm ) 0 with x Broken Parts Cost Quantity x · pm − cm x 0 1−x To derive the aggregation result, the probability distribution of x is required. Assume that each complete capital good consist of m parts and here m is normalized to be 1. By the end of each period, there will be at least m parts broken for sure, and the number of parts broken falls within [m, 1] stochastically with a probability measure, H : B([m, 1]) → [0, 1]. Unlike McGrattan and Schmitz (1999), Boucekkine and Ruiz-Tamarit (2003), Boucekkine et al. (2010), and Albonico et al. (2013), who directly assume the functional relation between depreciate rate of capital and maintenance expenditure, this paper makes assumption on the density (h) of H. To in line with the former literatures, this paper assume the following functional form: h(x) = ξ m , with ξ = and x ∈ [m, 1] 2 x 1−m The depreciate rate of capital goods are endogenously determined: Z pd −cm pm δd (pd , pm ) = 1 − dH(x) m 2 (2) (3) Let D be the capital goods stock held by CP’s. Then the quantity of parts demanded, M and the quantity of parts traded-in, R are as follows: Z M (pd , pm , D) = ρm (pd , pm ) · D = pd −cm pm xdH(x) · D (4) (1 − x)dH(x) · D (5) m Z R(pd , pm , D) = ρr (pd , pm ) · D = 2.2 m̄ pd −cm pm Investment Goods Producers There are two types of IP’s in the market; one is good and the other is bad. Bad IP’s suffer s fixed probability (q) of default at the beginning of each period due to high debt burden; while good IP’s do not. If one IP defaults on its debt, the firm will be forced to liquidate all capital and leave the market permanently. The default is assumed to be exogeneous in this simple model. Let j ∈ {g, b} denotes the type of IP’s, which is also known by CP’s. I assume there is no cost for IP”s to transfer from capital goods to parts or from parts to capital goods, which means capital goods and maintenance goods are perfect substituts in terms of production. Hence, sum of the parts an IP produces and gained from trade-in should equal the sum of new capital goods sold in the fore-market and maintenance goods sold in the after-market. According to the results in the previous section, once pd and pm are determined, both the quantity of maintenance goods demanded and the quantity of trade-in parts are determined. I assume the production technology is K α N 1−α ; thus, K α N 1−α is total number of parts that the IP produces; ρr (pd , pm )·D is the number of parts that CP’s have traded in. Therefore, the number of new capital goods sold is K α N 1−α − (ρm (pd , pm ) − ρr (pd , pm )) · D. For convenience, I denote Q(K, N, pd , pm , D) be the quantity of new capital goods sold in the fore market, hence: Q(K, N, pd , pm , D) = K α N 1−α − (ρm (pd , pm ) − ρr (pd , pm )) · D (6) State Variables, j ∈ {b, g} Kj Dj ξj Capital stock of IP’s Capital goods stock in market Indicator of default: 0 means default; 1 means not Action Variables, j ∈ {b, g} State transitions: if ξ j = 1, j ∈ {b, g} 0 K j = (1 − δk ) · K j + I j (7) 3 Ij Nj Pdj j Pm Investment of IP’s Labor hired by IP’s Price of new capital goods Price of maintenance goods 0 Dj = 1 − δd (pjd , pjm ) · Dj + Q(K j , N j , pjd , pjm , Dj ) (8) Reward function, dividend if ξ b = 1, j ∈ {b, g} fd (I j , N j , pjd , pjm , K j , Dj ) = γ · pjd · Q(K j , N j , pjd , pjm , Dj ) − w · N j − κ · K j − φk (I j , K j ) − I j + pjm · ρm (pjd , pjm ) · Dj (9) Here, w is exogenous wage rate; κ is proportional cost of production; and γ ∈ (0, 1) reflects the selling cost for new capital. φk is the adjustment costs for capital with functional form: φk (I, K) = ck · 2.3 I2 K (10) Consumption Goods Producers CP’s hold the capital goods produced by good and bad IP’s (Dcg and Dcb ). For CP’s, these are two types of capital. However, this simple model leaves the only difference between Dcg and Dcb be their producers ability to provide maintenance in the future. Since the policies of maintenance are already known, the only action that is undetermined is investment. State Variables, j ∈ {b, g} Dcj Capital produced by type j IP’s Action Variables: j ∈ {b, g} Ycj Investment bought from type j IP’s State transitions, j ∈ {b, g} 0 Dcj = (1 − δd (pjd , pjm )) · Dcj + Ycj (11) Here, δd (·) presents maintenance technology, defined in previous section. Reward function, profit fc ({Ycj , Dcj }j∈{b,g} ) = X Dcj − pjd · Ycj − pjm · ρm (pjd , pjm ) · Dcj j∈{b,g} (12) − cm · 1 − δd (pjd , pjm ) · Dcj Here, I assume CP’s adopt AK production technology. Let C be the quantity of consumption goods produced, then C = Dc . 4 2.4 Timeline Among the measure 1 IP’s, θ of them are good, 1 − θ of them are bad. At each period, CP’s utilize capital to produce consumption goods, and at the end of period T0 , some capital breaks out. For next period’s production, CP’s also have to make decisions on purchasing new capital and maintaining old capital. IP’s produce maintenance goods and new capital goods using their capital stock. I assume the capital goods produced by different IP’s are not compatible; thus, each IP actually form a small sub-market, and they monopolize their own sub-market. Therefore, the IP’s also decide prices to maximize their profit. In this simple model, there are only three periods: T0 , T1 and T2 . Period 0: The economy starts at T0 . The CP’s are endowed with D0b capital produced by bad IP’s and D0g capital produced by good IP’s. Good IP’s and bad IP’s are all endowed with K0 capital. And we assume IP’s and CP’s are initially ”n-to-1” paired. Period 1: At T1 , q of the bad IP’s will have to default on their debt and exit the economy for ever. The measure of bad IP’s now becomes (1−θ)(1−q). All the good IP’s will stay in the market. Period 2: At T2 , all IP’s exit the economy at the beginning of T2 , thus there is no market for capital goods. CP’s will keep producing consumption goods this period and exit the market at the end of T2 . 2.5 Maximization Problems Consumption goods producer Vc (D0b , D0g ) = max 1 X j {Yc,t } t=0,1 t=0 j∈{b,g} j j g b + Dc,2 ) β t · fc ({Yc,t , Dc,t }j∈{b,g} ) + β 2 · (Dc,2 (13) Subject to j j j Dc,t+1 = 1 − δd (pjd,t , pjm,t ) · Dc,t + Yc,t , t = 0, 1; j ∈ {b, g} (14) Investment goods producer: good Vd (K0g , D0g , g) = max {Itg ,Ntg ,pgd,t ,pgm,t }t=0,1 1 X β t · fd (Itg , Ntg , pgd,t , pgm,t , Ktg , Dtg ) (15) t=0 Subject to g Kt+1 = (1 − δk ) · Ktg + Itg , t = 0, 1 g Dt+1 = 1 − δd (pgd,t , pgm,t ) · Dtg (16) + Q(Itg , Ntg , pgd,t , pgm,t , Ktg , Dtg ), 5 t = 0, 1 (17) Investment goods producer: bad Vd (K0b , D0b , b) = 1 X max {Itb ,Ntb ,pbd,t ,pbm,t }t=0,1 β t (1 − q)t · fd (Itb , Ntb , pbd,t , pbm,t , Ktb , Dtb ) (18) t=0 Subject to b Kt+1 = (1 − δk ) · Ktb + Itb , t = 0, 1 b Dt+1 = 1 − δd (pbd,t , pbm,t ) · Dtb (19) + Q(Itb , Ntb , pbd,t , pbm,t , Ktb , Dtb ) 2.6 t = 0, 1 (20) Market Equilibrium The CP’s and IP’s maximize their value functions according to the maximization problems described above given the initial values. Both fore-market and after-market are clear. Due to the Walras’s Law, we only require: α j Yc,t = (Ktj ) (Ntj ) 1−α − ρm (pjd,t , pjm,t ) − ρr (pjd,t , pjm,t ) · Dtj , j ∈ {b, g}, t = 0, 1 (21) 3 Analytical Solutions The central task of solving this model is to solve the prices pm and pd . According to the results in the maintenance section, it is equivalent to consider pd and m∗ , rather than pd and pm , in solving this simple model. The reason for using m∗ is that it provides convenience in taking derivatives. I solve the prices using backwards induction, starting with prices in T1 . Since, all IP’s will exit the economy at T2 , there is no difference between the good and the bad, and no IP’s will make any investment at T1 , All IP’s will focus on the dividends at T1 . Hence, the following FOC’s holds: I1j : 0 = I1j (22) R mj mj1 : 0 = −γ · pjd,1 h(mj1 ) N1j : 0 = (1 − α)γ · pjd,1 + (pjd,1 (K1j )α (N1j )α 1 − cm ) h(mj1 ) − −w m xdH(x) (mj1 )2 (23) (24) Given that the CP’s also know that all IP’s will exit at T2 , the price of new capital goods in T1 can be determined: pjd,1 = β (25) 6 Therefore, if interior solution can be obtained, mj1 and pjm,1 should be: mj1 = e (1−γ)β−cm +ln(m) β−cm (26) β − cm pjm,1 = e (27) (1−γ)β−cm +ln(m) β−cm Next, I calculate the prices of good IP’s at T0 . Since the good IP’s will stay in the economy at T1 , the capital goods bought from them at T0 can be used for more than one period given proper maintenance. Similar to what has been modeled in other durable goods literatures, CP’s evaluation now is also related to expected prices of capital goods and maintenance goods at T1 , which can be clearly seen in the FOC’s of CP’s problem: g Yc,0 : pgd,0 = β · 1 − cm · 1 − δd (pgd,1 , pgm,1 ) − pgm,1 · ρm (pgd,1 , pgm,1 ) + β 2 · 1 − δd (pgd,1 , pgm,1 ) (28) Given the results in previous sections, the forward-looking pricing formula can be written explicitly: ! Z mg Z mg 1 1 1 g pd,0 = β + β(β − cm ) dH(x) − g · xdH(x) (29) m1 m m Given the results of T1 , pgd,0 is fully determined by parameters. For other actions, the behaviors of IP’s are fully captured by the FOC’s: I0g g 1−α g α−1 g ) − κ ) (N (K + β αγ · p 1 1 d,1 K0g R mg 0 xdH(x) m mg0 : 0 = −γ · pgd,0 h(mg1 ) + (pgd,0 − cm ) h(mg0 ) − (mg0 )2 ! g α ∂fd,1 (1 − α)γ · (K0g ) g g N0 : 0 = pd,0 + β · −w α ∂D1g (N0g ) I0g : 0 = −2ck · (30) (31) (32) where: g ∂fd,1 ∂D1g = ∂fd (I1g , N1g , pgd,1 , pgm,1 , K1g , D1g ) ∂D1g = γpgd,1 Z m̄ mg1 (1 − x)dH(x) + β − cm − γpgd,1 mg1 Z mg1 xdH(x) (33) m Note that if a capital goods can only be used for one period, the price should be β. Therefore, given the probability of bad IP’s default at T1 , the pricing formula for bad IP’s can be written as follows: ! Z mb Z mb 1 1 1 b dH(x) − b · xdH(x) (34) pd,0 = β + β(1 − q)(β − cm ) m1 m m 7 Therefore, the FOC’s of bad IP’s can be derived as follows: b α−1 I0b b (K1 ) : 0 = −2ck · b + (1 − q)β αγ · pd,1 b α−1 − κ K0 (N1 ) R mg 0 xdH(x) m mb0 : 0 = −γ · pgd,0 h(mg1 ) + (pgd,0 − cm ) h(mg0 ) − (mg0 )2 ! α b ∂fd,1 (1 − α)γ · (K0b ) b b pd,0 + (1 − q)β · N0 : 0 = −w α ∂D1g (N0b ) I0b (35) (36) (37) where: b ∂fd,1 ∂D1b = ∂fd (I1b , N1b , pbd,1 , pbm,1 , K1g b, D1b ) ∂D1g = γpbd,1 Z m̄ (1 − x)dH(x) + mb1 β − cm − γpgd,1 b m1 Z mb1 xdH(x) (38) m Note that (δk · k)2 k (x − ζ · b)2 φb (b, x) = cb b 1 1−α2 w , with γ0 = 1, γ1 = 1.1 n(k, i) = γi α α2 · p · k 1 φk (k, i) = i · ck (39) (40) (41) Here, φk and φb are the adjustment costs for capital and debt, respectively, and n is labor. CP’s D Capital of CP’s K D ξ IP’s Capital of IP’s Capital goods stock in market Indicator of default Action Variables CP’s Y M Investment of CP’s Maintenance goods Yd I Pd Pm IP’s New capital goods sold Investment of IP’s Price of new capital goods Price of maintenance goods State transitions if i = 1 k 0 = (1 − δk )k + δk k (42) d0 = k α1 n(k, i)α2 + (1 − δd )d (43) 8 b0 = x (44) Here, n denotes labor. Reward function, dividend f (k, d, b, x, i) = p · k α1 n(k, i)α2 − κ1 · k − κ0 − w · n(k, i) − pm · d − φk (k, i) − φb (b, x) − pb · b + qi · x (45) where (δk · k)2 k (x − ζ · b)2 φb (b, x) = cb b 1 1−α2 w , with γ0 = 1, γ1 = 1.1 n(k, i) = γi α 1 α2 · p · k φk (k, i) = i · ck (46) (47) (48) Here, φk and φb are the adjustment costs for capital and debt, respectively, and n is labor. Bellman equation V (k, d, b) = max f (k, d, b, x, i) + iβV (k 0 , d0 , b0 ) (49) x≥0 i∈{0,1} Here, V (k, d, b) denotes the maximum attainable present value of current and future dividends, given that, at the beginning of the period, the firm is in good standing with lenders and possesses capital stock k, durable stock d, and debt b. The value function if the firm is not in good standing with lenders, because it has defaulted, is, of course 0. Parameters 9 Symbol p w α1 α2 κ1 κ0 pm pb ck δk δd cb ζ β γ0 γ1 q0 q1 Value 1.183 1.5 0.6 0.3 0.55 0.05 0.02 0.9892 7.5 0.1 0.2 0.06 0.6 0.964 1.0 1.1 0.2 β Definition price of durable good, calculated in other part of the project wage rate share of capital in durable production share of labor in durable production fixed cost of production proportional cost of production maintenance profit rate debt payment, due to the tax shield, it’s between β and 1 capital adjustment cost parameter capital depreciation rate durable goods depreciation rate debt adjustment cost paprameter zero adjustment cost debt policy discount factor labor coefficient, if default next period labor coefficient, if repay next period price of new debt, default price of new debt, repay Given the formulation above, it appears to me that you have incorrectly titled κ0 and κ1 . κ0 is a constant term in your reward function, and thus indicates a fixed cost. κ1 appears to be a unit cost of capital maintenance or something related, yielding a total cost this is proportional to capital stock. Please revise the names in the table above as needed. Also, the name you give to pb is not very clear. It appears to be some kind of rate; if so, revise the name to this variable in the table above. 4 Numerical Solution We will compute an approximation to the value function via the method of collocation. Specifically, we approximate the value function using a linear combination of N three-dimensional Chebychev polynomials φj : X V (k, d, b) = cj φj (k, d, b). (50) j We will then chose a series of N nodes (kj , dj , bj ) and compute the unknown coefficients cj by solving the N nonlinear equations simultaneously X X cj 0 φj 0 (kj , dj , bj ) = max f (kj , dj , bj , x, j) + iβ cj 0 φj 0 (k 0 , d0 , b0 ) . (51) x≥0 0 0 j j i∈{0,1} j = 1, 2, . . . , N I am going to propose a derivative-free algorithm to solve the collocation equation that should prove to be fast and stable. However, I am not going to explain the algorithm now. We first need to get on the same page about the model. 10 5 Questions 1. When I substituted your investment policy equation into the capital transition equation, the latter reduced to k 0 = k (see equation 1). This makes no sense. To simplified the exogenous investment policy, I assume fixed investment rate: i = ξ · k. ξ can be other numbers. I don’t understand you answer. You have not changed equation (1). It still reads k 0 = k. This means that k is fixed and fully exogenous to the model, and thus not a true state variable. The model cannot be solved unless you specify its value, and you have not done so. I would like to proceed to advise you on how to solve the model numerically. However, I cannot do so until you resolve how you want to treat k. If it is truly unchanging, then tell me what its value is, and revise the model formulation above to exclude any mention of k as a state variable. 2. When I substituted your investment policy equation into the capital adjustment cost equation, it reduced to a linear function of k (see equation 5). Is this what you intended? No, this is true only when fixed investment rate is assumed. When we fully solve the model, investment policy should have some curvature. I also don’t fully understand you answer. You have not changed equation (5) above. If you want to proceed with equation (5) as written, fine ... just confirm that this is the case. If not, then edit equation (5) to what you believe it should say. 3. You state then when the firm decides to default the following period, it will announce this publicly, making it known to lenders. You then reduce the cost of capital q from β ≈ 0.9 to 0.2. Why would the firm’s cost of capital fall if it publicly announces an intent to default the following period? Indeed, why would any lender lend any amount to the firm, given that it has announced its intent to default. The price of debt q falls because the return of the debt falls when firm default. If the firm declares default, then creditors will liquidate the firm’s capital. If the liquidation value is 0, then q should be zero; if the liquidation value is positive, then q > 0. Ideally, the liquidation value should be a function of capital stock, but I failed to solve the model with this assumption. Therefore, I assume q = 0.2 instead. Also, here I assume that the financial intermediaries has full information, thus they know the expected return of firm’s debt and can price the debt rationally. It is different from ”announce to default the following period”. Later, I will introduce households who do not have such information. I will not question your reasoning on this matter further, and continue to treat q as it is specified above. 6 Comments 1. I am unwilling to let i denote the transition indicator, since eventually I will have to solve the investment. and i is commonly used to denote investment in macroeconomics literatures. 11 I understand that the intended audience for your research will be most comfortable with i used to indicate investment. However, as the subtitle “Implementation Notes” above conveys, this document is designed to guide us in developing working numerical code. In numerical work, i is universally used as a counter or discrete variable, and in developing working code for your model, we are going to use it to represent the default choice. If, as proceed, you reintroduce investment to the model (which for now has disappeared due to substitution) you will need to find a different variable name for it. However, this does not commit you to using that variable name in the narrative of your dissertation or working paper. You can, in your narratives, use i to indicate investment. 2. We can first solve the simpler version with the assumption qi = q̄. But in the baseline model, qi is indeed a function of i since firms’ future default decisions can be rationally expected by the lenders. And this makes the model more challenging and interesting. I am not sure what to make of this comment. Keep in mind that all we want to do right now is to specify a very basic version of model that retains only the most essential features of the model you ultimately wish to solve, so we can proceed to developing working code. Once we have developed working code for the simple model, we can begin to introduce more complicated features. 3. For the derivative-free method, I have tried the N elder − M ead method to solve the optimal decisions at each grid node. I found it is not stable. Is the method in your mind related to Nelder-Mead method? No. The derivative-free method I am going to propose for solving your model has nothing to do with Nelder-Meade. This will all become clear later, once we are ready to develop initial working code, at which time I will explain the matter further. However, we cannot proceed to writing numerical code until we have a viable, simplified, fully parameterized model to work with. And, as the discussion above makes clear, we have not yet reached this point. Finally, you have sent me some of your code, but I do not know what you hope to have me do with it. The code is not even minimally documented, and neither have you provided side notes (like the implementation notes in this document) that explains what the code is supposed to do. Moreover, the parameters you specify in your code do not conform to those used in this document. You must develop discipline when you write code. The code must documented sufficiently to give the reader some sense of what each code segment is supposed to do. Moreover, your code should be accompanied by side notes, like these implementation notes, that describe what the code is designed to do more clearly. Keep in mind that writing implementation notes to accompany your code will not be wasted work. In your dissertation and working papers, you will need to explain your numerical procedures, and most of the language in your implementation notes will have to appear in your paper, albeit perhaps in abbreviated form or as an appendix. Your implementation notes must be sufficiently clear and detailed for me to fully understand what the code is designed to do. If I cannot understand what are doing, readers of your dissertation and papers certainly not be able to understand. In any case, I am ignoring the code you sent me, not only because you have failed to document it properly, but because I cannot see how it can help us solve the model numerically. 12