A Macroeconomic Model of Endogenous Systemic Risk Taking D. Martinez-Miera and J. Suarez Discussion Rafal Raciborski DG ECFIN, European Commission Norges Bank, Oslo, 29 - 30 November 2012 Disclaimer The views expressed are the author’s alone and do not necessarily correspond to those of the European Commission. Context • It's been almost 5 years that the world has been in the financial and economic crisis… • …with its causes still not yet fully understood… • …but with a contribution of the financial sector generally unquestioned Most economists would agree the financial sector (banks in particular) may contribute to and perhaps generate systemic risk This paper • Discusses one particular channel via which systemic risk may originate in the banking sector – Idea most closely linked to the 'risk-shifting literature’ • Embeds it into a general equilibrium model – May be disputed whether the systemic risk is truly endogenous; more on it later • Solves nonlinearly to discuss optimal bank capital requirements The model: general idea • General result (Jensen&Meckling, 1976; Stiglitz&Weiss, 1981; Allen&Gale, 2000): – Limited liability βΉnon-convexities in the profit maximizer's problem – The maximizer may then prefer a riskier project, pushing its risk on other agents (=risk shifting) • Banks protected by deposit insurance (οlimited liability)βΉ they like riskier projects • But: riskier behaviour≠systemic risk – Assume that riskier projects are systematically linked The model: available projects • 2 types of projects: 1. Less risky projects (in terms of its variance and its mean): idiosyncratic risk 2. More risky projects: risk perfectly correlated • Higher variance of the risky projects to induce risk-shifting in the banks • Correlation of risky projects=systemic risk • Lower unconditional mean of the risky project probably makes things harder; conveys the idea of systemic risk being "bad" The model: equilibrating force Due to limited liability banks like riskier projects; why don't we observe only the riskier ones being chosen (share of risky projects x=1)? • Crucial variable: stochastic marginal value ππ‘+1 (ππ‘ ) of one unit of a banker's wealth • Upon the realization of the systemic risk: – Wealth of 'risky banks' is wiped out – Scarce ππ‘ ⇒ ππ‘+1 driven up for save banks: last bank standing effect (in the spirit of Perotti&Suarez, 2002) • In equilibrium banks indifferent between projects βΉ x∈ (0,1) Welfare • Banks’ agency problem affects negatively the economy via 2 channels: – Static losses: picking inefficient projects – Dynamic losses: loss of bank equity (and, hence, lending capacity) in the event of a systemic shock • Measurement: – All agents risk neutral; but GDP does not reflect welfare well – GDP (=added value) excludes capital losses – Does output (y=GDP+undepreciated K) correlate perfectly with welfare in your model? Capital requirements • Increased capital requirements γ make capital scarcer (βΉ ππ‘+1 higher) βΉ higher incentive to choose safer projects βΉ higher proportion of bank equity invested in safer projects • But, banks’ lending capacity reduced βΉ lower average efficiency • Trade-off βΉ optimal γ ∈ (0,1) Results • For the benchmark calibration: – With low γ=7% fraction of capital invested in systemic projects very large (70%) βΉ – Systemic shocks very painful (31% drop of GDP) – Optimal γ large (14%) – Optimal γ βΉ welfare higher by about 1% • Number of extensions – Interesting perverse results Minor remarks (I) • You assume a pooling equilibrium – Are there other types of equilibria? – If so, how do we know yours is the relevant one? • One of your main contributions: quantitative results (“high optimal γ”); but your model ‘very stylized’. For example: – Crucial role of the slope of ππ‘+1 (ππ‘ ) – It would be less steep if labour were variable… Minor remarks (II) • An issue with calibration? – You assume 35% depreciation in failed firms – For γ=7%, 70% of all projects are systemic – This gives 35%×70%=25% capital depreciation in the economy in the event of a systemic shock – Also the fall in GDP (30%) very large • Develop the sensitivity analysis – “The choices for the values of […] ψ and φ are quite tentative.” General equilibrium? Is systemic risk endogenous? • Yes: share of bad projects x=f(π,regulation) • No: systemically-risky projects are always there to be picked βΉ only the severity of the crisis endogenous I believe we cannot do w/o opening the black box – see next 2 slides Take the black box as given What are the systemic projects? • Allen&Gale (2000): oil shock – convincing, but with a limited application (Norway!) • Your footnote 1: housing bust: – Is it systemic? What makes it so? – Was it (before 2007) considered risky? (The notion that “house prices never fall”) • Even so: Is it plausible? Convince the reader! • What happens in your model if you have 2 types of risky projects: identical payoffs, but projects of the 2nd type independent Bring your channel to the data “Systemic Banking Crises facts” (Boissay et al.): a) SBC’s are rare and deep b) SBC’s are closely linked to credit developments Ad. a) Your model can obviously match it, but: – by imposing exogenous prob. of a systemic crisis – endogenous risk correlation in recessions, Brunnermeier&Sannikov, 2011 (parsimony) Ad. b) Nothing to say about it – again, endogenous link (Boissay et al., 2012) – hard to make policy advice w/o a crucial channel Need to open up the black box Interesting perverse effect? • Your results sensitive to the exogenous probability of a systemic crisis – Benchmark: ε=0.03 • One view: makes your results fragile • Alternative view: innovations that make the economy safer (εβ) make crises deeper… Worth exploring?