Liquidity Regulation and Risk of Runs: Last Taxi Must Stay

1. “Liquidity regulation and the risk of runs: Why the last taxicab must NEVER leave the train station” Douglas W. Diamond Chicago Booth and NBER and Anil K Kashyap Chicago Booth and NBER Becker Brown Bag November 21, 2016

2. Basel III and Liquidity Regulation • What is the goal of liquidity regulation? • Why won’t banks hold the proper amount otherwise? • Is more disclosure a better answer?

3. Liquidity requirements in Basel III • The Net Stable Funding Ratio: • (Type of funding tied to assets) Ties the liability structure to the liquidity characteristics of assets. Liabilities are assumed to have varying “stability” given their maturity (based on counterparty, core deposits, etc.) Measured over one year horizon. Can be violated for a period. • The Liquidity Coverage Ratio (liquidity minwhich must hold at all times):

4. Are Liquidity Requirements a Joke?Require Liquidity which can’t be used. • Goodhart (2008) analogy. Last Taxi at the train station. • Are regulators crazy?

5. “The weary traveler who arrives at the railway station late at night, and, to his delight, sees a taxi there who could take him to his distant destination. He hails the taxi, but the taxi driver replies that he cannot take him, since local bylaws require that there must always be one taxi standing ready at the station.”

6. Our perspective • Most analysis of liquidity requirements asks: how much liquidity is needed to meet extreme withdrawals (as in a crisis)? • Rather, a key goal should be to provide incentives for banks to chose to hold the proper amount of liquidity, in excess of the required amount. • This extra liquidity is to deter runs. • Unregulated banks may not hold enough liquidity. • Regulation that forces banks to hold more liquidity than they prefer can potentially improve outcomes.

7. It’s no joke: Back to Our Perspective on Liquidity Regulation 2. A key goal should be to provide incentives for banks to chose to hold the proper amount of liquidity, in excess of the required amount. (In order to deter runs). (Some is unused.)

8. Banks, Liquidity Creation and RunsDiamond-Dybvig (1983) • Banks and (shadow banks) create liquid liabilities (short term deposits) while holding illiquid assets (e. g. long- term loans). • It works when there are no runs, but this maturity and liquidity mismatch leaves them subject to runs (panics or fear of fear itself).

9. Basic Idea from Diamond-Dybvig (1983) • Banks deposits are more “liquid” than the loans that they hold • Very liquid means a very small loss for an early withdrawal of the deposit • Investors like liquid assets because they don’t know how soon they will need their money. • Banks can create liquid deposits out of illiquid loans.

10. Investor Demand for Liquidity • Investors face an uncertain need for liquidity. Each will need their money (to consume) either early at date 1 or late at date 2, and does not know which date as of date 0. Each begins with $1 to invest on date 0 . • If bank will be sure to be solvent all the time, only those who are early will withdraw at date 1.

11. Available Assets:Illiquid Loans • T=0 T=1 T=2 -1 θR2<1 0 (liquidate early) or 0R2>1 (hold to maturity) Liquid Assets T=0 T=1 T=2 -1 R1> 10 or 0 R1* R1(roll over)

12. Can the bank Create Liquid out of Illiquid? • Can bank pay 1 on demand to each depsoitor who deposits 1 at date 0 or 1? • Yes, if only some depositors (early ones) show up at date 1. • No, if all or most depositors show up (or are expected to show up) at date 1? Bank is subject to runs. • Can the bank survive a “partial run” where some depositors MIGHT panic?

13. How many will need liquidity? • Suppose a fraction ts<1 of depositors will be early (need money at date 1) and 1-ts late (at date 2). • To be concrete, suppose today there are 100 such investors and ts=0.25, the 25 will be early, and 75 will be late. • It is not known which investors will be early or late. As of date 0, each investor has a 25% chance of being early, and 75% chance of being late. • At other times, ts will vary (e.g,. ts=0.4).

14. Assets and Deposits • 1 invested at date 0 in: at date 1 pays at date 2 pays • Liquid asset R1 > 1 R1*R1 • Loan asset θR2<1 or R2>R1*R1 • Deposit r1=1r2= 1 Runnable if a sufficient fraction is in Loans. General case in the paper has r1 and r2 not necessarily equal to 1.

15. t=0 Riskless loan and safe assets are funded by deposits t=1 Idiosyncratic need for liquidity t=2 Loans repaid Our Variant of Diamond-Dybvig (1983) Fraction ts fundamentally must withdraw Δ of patient people see a sunspot Investment payoff Bank run possible Liquidated loans repay θR2

16. Details on deposits withdrawals and Partial Runs • A fraction tsof depositors will need to withdraw, ts varies and is known only by the bank. • In addition, a fraction Δ<1 of depositors see a report that can make them expect others to run (see a “sunspot” or some news). This may or may not make them run in response. • Δ measures how “hot” is the money deposited. Core deposits: Δ=0 vs. Wholesale deposits: Δ>0. • Runs are partial: ts +Δ<1.

17. To be run free, bank must hold some extra liquidity • Enough liquidity so if a fraction ts+Δ show up, the bank will still be solvent. • If the bank holds the extra amount, and all know it, there will not be runs. • Once there are no runs, this is unused liquidity and will be unused (extra taxicabs at the train station) • But if no extra liquidity, there can be runs.

18. Why do we need to provide incentives and not just disclosure? • Disclosure need not allow depositors to determine if there is “enough liquidity.” • Disclosed numbers are difficult to interpret because: • Bank has information about varying needs for liquidity. • A snapshot on a date my be stale (hold liquidity on December 31, invest it the next day). • Disclosing temporarily low liquidity could cause a run on its own.

20. Bao, David and Han (2015)

21. Privately optimal choices for the bank • Fraction of liquid assets (α) is chosen to equate available funds (α) to deposit outflows (1 each to the fraction f1 that withdraw: (total outflow of f1). • This profit maximizing amount is “Automatically Incentive Compatible.” (AIC) • Because the bank never plans to make illiquid loans only to liquidate them at a loss. • Withdrawals differ: without a run, f1=ts, or in a run, f1= ts+Δ .

22. Automatically Incentive Compatible Liquidity, for given f1 withdrawals. α= Fraction in liquid asset

23. Automatically Incentive Compatible Liquidity, for given f1 withdrawals. α= Fraction in liquid asset tlow

24. Automatically Incentive Compatible Liquidity, for given f1 withdrawals. α= Fraction in liquid asset thigh

25. Withdrawals differ without a run, f1=ts, and in a run, f1= ts+Δ . α= Fraction in liquid asset ∆ ts ts+∆

26. Will bank choices deter runs? • If the bank can cover withdrawals of f1 in all cases without failing, the hot money never runs. • Will bank choose enough liquidity for normal withdrawals f1=ts, orto stay solvent even during a run, f1= ts+Δ?

27. Definitely deterring runs can require more liquidity than self-interest assures (With Full Information about ts) α= Fraction in liquid asset

28. Stability Requires Some Unused Liquidity (With Full Information about ts) α= Fraction in liquid asset

29. Does simply requiring excess liquidity overcome private information about • Not generally: • When there is full information, all liquidity is released in any run ( ) (all taxicabs can leave the train station). • If a regulator does not know , releasing this liquidity only in a run may not be feasible.

30. Not all liquidity (taxicabs) can be released in all runs (if is not known) α= Fraction in liquid asset ∆

31. Evaluation of Basel III regulations • We can show that the Basel regulations are NOT optimal regulations (constrained only by requiring honest reporting of the bank’s private information). • They require more liquidity and less lending than the optimal mechanism. • We can still compare them using our framework.

32. Liquidity requirements in Basel III • The Net Stable Funding Ratio: • (Type of funding tied to assets) Ties the liability structure to the liquidity characteristics of assets. Liabilities are assumed to have varying “stability” given their maturity (based on counterparty, core deposits, etc.) Measured over one year horizon. Can be violated for a period. • The Liquidity Coverage Ratio (liquidity minwhich must hold at all times):

33. Run-Proof NSFR Must Cover the Worst Case α= Fraction in liquid asset

34. LCR [of ρ(1-f1)] can allow more lending that a NSFR (α) α= Fraction in liquid asset

35. LCR [of ρ(1-f1)] can allow more lending that a NSFR (α) α= Fraction in liquid asset Liquid holdings under LCR (selected by bank).

36. There must be excess liquidity • To enforce the LCR regulation, the regulator need only measure how much liquidity per deposit remains after withdrawals occur. • There must be a positive fraction of liquidity left unused after fraction ts withdraw. Last taxicab at the train station must not leave. • Regulator can’t tell if withdrawals are normal or run, but if the extra liquidity is held, only normal withdrawals will occur.

37. Implementing the optimal regulation(why we can do better) • The best regulation (better than the LCR) be implemented by lender of last resort policy tied to the full information unused liquidity requirement . • If violated (by using it to meet a run), lend against the liquidity, but drive banker compensation to zero. • A penalty for the last taxi leaving the station! • As in dividend prohibition rules the original Federal Reserve Act.

38. Summary • Unregulated banks with unobservable liquidity needs are unlikely to be run proof. • Simply disclosing liquidity at one date is not enough. • Liquidity regulation can correct this. • Basel style regulations are not the optimal mechanisms. They will typically result in excess liquidity being held. • Mandating surplus liquidity is necessary, so the last taxicab will remain at the station. • Lender of last resort policies and liquidity regulation ought to be integrated (allows few taxicabs to be wasted).

40. The recent crisis is like all others • Private financial crises are everywhere and always due to problems of short-term debt (and to the reasons why short-term debt is needed). • This is about runs. • The trick is to find the short-term debt (sometimes it is well hidden).

41. Find the Short-term “debt”= Runnable Claims • Third Avenue Focused Credit Mutual Fund (2015). • Institutional Money Market Funds (2008). • Lehman Commercial Paper and Repo. • AIG Collateral Calls. • East Asia, 1997: short-term foreign debt. • LTCM/ Russia 1998: Margin loans of LTCM. • Stock Market Crash/ Reversal 1987. • Delays in order processing prevented real time orders and information • Bank runs in 1930’s. • Demand deposits.

42. The following slides are background data and extensions which I will not discuss except possibly during the Q&A.

43. Source: http://www.rba.gov.au/publications/fsr/2012/sep/html/tables.html#table-a2

44. Bao, David and Han (2015)

46. Measurement and Calibration Issues • The illiquidity, θR2, can be higher of market or LLR (lender of last resort after a haircut). • We should not calibrate liquidity requirements just to cover predicted withdrawals, but in instead take account of the incentive effects of requiring unused liquidity (LCR). • Behavior in the near future will be very different with requirements to hold liquid assets with higher interest paid on reserves by central banks (set to induce holdings of reserves).

47. What about Capital Requirements alone? • Require more capital (used to finance more assets / loans) per unit of deposits. • Works well if assets are reasonably liquid (θ is large, loans serve as collateral against a run). • Works poorly if loans are very illiquid: if θ=0, adding capital and more loans has no reduction in the threat of runs.