Liquidity Crisis, Runs, and Security Design

Liquidity Crisis, Runs, and Security Design Lessons from the Collapse of the Auction Rate Securities Market Song Han and Dan Li Federal Reserve Board Chicago, May 7, 2009 The views presented herein are completely our own and do not necessarily reflect those of the Board of Governors of the Federal Reserve System. Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 1

Main Results ARS design is flawed Vulnerability to two runs: Investers run, partially panic-based Broker-dealers run Unexpected first mover withdrawal of liquidity support triggered simultaneous withdrawal by all major broker-dealers Two runs interact and amplify each other Problems with uniform price auction Strong evidence of underpricing Auction reset rates only weakly related to fundamentals, positively related to maximum rate Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 2

Auction Rate Securities (1984-2008) Long-term debt securities with variable interest rates, reset every 7, 28, 35 days etc. through a Uniform-Price Auction Process Purpose: ARS allows issuer to fund long term liability with short term debt Issuers: municipalities, close-end funds and student loans authorities Investors: corporate treasury, high net-wealth individuals Market size: $330 billion (end of 2007), half in Muni ARS (MARS) ARMB Issuance and Interest Rates Amount in billion dollars 0 5 10 15 20 Issuance in billion dollars (left) SIFMA swap rates Bond Buyer GO Bond Yields 0 2 4 6 8 Percent 1990q1 1995q1 2000q1 2005q1 2010q1 Quarter Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 3

Auction Process & Dealer s Role Purpose Order Types Price discovery: Setting interest rates Existing Owner : Sell, Hold, Hold at rate Source of liquidity: transfer ownership Potential Buyer : Buy at rate Clearing Rate: the lowest rate at which bids are sufficient to cover all sells In the example, the clearing rate is 6% Rate 10% Maximum rate=10% A 6% 5% Clearing rate=6% 3% Demand Curve Supply Curve Units 10 If sells > buys, auction fails. Transfer prorated. Reset rates set at maximum rate Dealer can participate after seeing the demand curve to support auctions, but not required to do so. Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 4

Auction Process & Dealer s Role Purpose Order Types Price discovery: Setting interest rates Existing Owner : Sell, Hold, Hold at rate Source of liquidity: transfer ownership Potential Buyer : Buy at rate Clearing Rate: the lowest rate at which bids are sufficient to cover all sells In the example, the clearing rate is 6% Rate 10% 6% Maximum rate=10% Clearing rate=6% A Rate 10% 6% Maximum rate=10% Fail A Reset Rate 5% 5% 3% Supply Curve 3% Demand Curve Units Demand Curve Supply Curve Units 10 If sells > buys, auction fails. Transfer prorated. Reset rates set at maximum rate Dealer can participate after seeing the demand curve to support auctions, but not required to do so. 10 Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 4

Auction Process & Dealer s Role Purpose Order Types Price discovery: Setting interest rates Existing Owner : Sell, Hold, Hold at rate Source of liquidity: transfer ownership Potential Buyer : Buy at rate Clearing Rate: the lowest rate at which bids are sufficient to cover all sells In the example, the clearing rate is 6% Rate 10% Maximum rate=10% A Rate 10% Maximum rate=10% A 6% 5% Clearing rate=6% 6% 5% Clearing rate=5% 3% Supply Curve 3% Dealer s bid Demand Curve Units Demand Curve Supply Curve Units 10 If sells > buys, auction fails. Transfer prorated. Reset rates set at maximum rate Dealer can participate after seeing the demand curve to support auctions, but not required to do so. 10 Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 4

Before Week of Feb 12: Managed Bidding Dealer s impact in: Pricing: Price talk and Actual bids Liquidity: net Buyer in auctions, net Seller in non-auction secondary market False Sense of Safety Many investors are unaware of auction dealer s role in auctions Rates of Auction Failure on ARS Failure rate 0.2.4.6.8 Actual failure Pseudo failure 01jul2007 01oct2007 01jan2008 01apr2008 Date Note. Based on data from three major auction agents Many insiders expect the implicit support to be binding Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 5

Data in MARS Auction results from three main auction agents Auction status, reset rates, benchmark index rate Muni transactions data from MSRB Trade size, price, direction of trade Bond characteristics from Bloomberg Identify maximum rates through rule matching Three sample periods Pre-crisis period 7/1/2007-12/31/2007 Crisis period 2/11/2008-2/19/2008 Post-crisis period 2/20/2008-3/19/2008 Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 6

Empirical 1: Investors Run Unusually large number of sell orders on Feb 12 0 10 20 30 40 Successful Auctions Failed Auctions Feb 12 Par_sell Jul Sep Nov Jan Mar May 2007 2008 Sudden surge in the incidence of failures (Pseudo Fail+Actual Fail) Failure rate 0.2.4.6.8 1 Actual failure Pseudo failure Order imbalance 12Feb08 01Jul07 01Oct07 01Jan08 01Apr08 Note. Based on data from three major auction agents Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 7

Determinants of Auction Failures Fundamental variables: bond characteristics, credit risk, macro factors Max rate: auction likely to succeed if maxrate upper support of fundamental value Failure Rate 0.2.4.6.8 1 Failure of high max rate bonds Failure of low max rate bonds Pseudo failure of high max rate bonds Pseudo failure of low max rate bonds 12Feb08 01jul2007 01oct2007 01jan2008 01apr2008 Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 8

Abnormal Failure Rates Panic based (Sunspot) v.s. Informational based? Abnormal Failure Rates in Mid-February Predicted Abnormal Std. Dev. t-stat of Date Actual p t ˆp t p t of pit N t p t 2/11/2008 0.04 0.42-0.40 0.44 225-13.64 2/12/2008 0.13 0.39-0.28 0.51 358-10.45 2/13/2008 0.60 0.43 0.11 0.48 385 4.49 2/14/2008 0.57 0.43 0.09 0.37 309 4.04 2/15/2008 0.57 0.38 0.11 0.32 359 6.79 2/19/2008 0.53 0.45 0.01 0.32 403 0.83 The runs are partially panic driven. Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 9

Dealer s Inventory Stress Broker Dealer Net Buys in and post Auctions Amount in Millions Dollars 4 2 0 2 4 6 Dealer net buy in auctions Dealer net buy post auctions 12Feb08 01Jan07 01Apr07 01Jul07 01Oct07 01Jan08 01Apr08 Cumulative Inventory Aggregated over All Broker Dealers Cum Inventory in Billion Dollars 20 15 10 5 0 12Feb08 01Jan07 01Apr07 01Jul07 01Oct07 01Jan08 01Apr08 Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 10

Dealers Run Failure to Coordinate Pct of Failed Auction by Dealer Around Crisis 0.8 Pct of Failure 0.6 0.4 BOFA CITI GOLD JPMO LEHM MERR OTHER UBS 0.2 2008 02 08 2008 02 11 2008 02 12 2008 02 13 2008 02 14 2008 02 15 One bank let their auctions fail, all others followed the next day Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 11

Why Simultaneous Withdrawal of Liquidity Support? One dealer s decision to Support or Not: Cost of support: Inventory cost, balance sheet stress Benefit: reputation to both investors and issuers Multiple Dealers: externality of one dealer s decision on others by letting auctions fail Investors run away from all ARS, other dealers forced to take more inventory, more stress to balance sheet Relative cost to reputation diminishes for other dealers if they also withdraw Two Equilibrium outcomes: All support (unstable), or all withdraw (stable) Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 12

Empirical 2: Uniform Price Auction Inefficiencies All bidders pay at the clearing rate Theoretical Predictions, Back & Zender 93, 01 1 Without dealer support fixed quantity auction Equilibrium may be unrelated to fundamentals Equilibrium with lowest price (highest interest rate) preferred 2 With dealer support endogenous quantity auction Difference between worse-case equilibrium and fundamental value converge to zero as number of bidder increases Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 13

Results from OLS Regressions of Reset Rates 7/1/07-12/31/07 2/20/08-3/19/08 Independent var. (1) (5) (6) (10) Maximum rate 0.012** 0.228** (0.00) (0.03) Lag. cum. inventory 0.050** 0.224** (0.01) (0.11) Lag. non-auc. trade -0.011 0.579** (0.01) (0.10) Bond, credit, macro Yes Yes Yes Yes R 2 0.66 0.66 0.21 0.30 N 34369 34369 3496 3496 In the pre-crisis equilibrium (endogenous supply model): reset rates reflect strongly fundamentals; auction variables such as the maximum rate are not relevant; In the post-crisis equilibrium (fixed supply model): reset rates are weakly related to bonds fundamentals; reset rates are positively related to maximum rate; Reset rates in the post-crisis equilibrium are increasing in the secondary non-auction market liquidity. (Less competitive bidding) Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 14

Summary of Findings and Lessons Findings ARS is flawed ARS crisis caused by two types of runs Prices in auctions can deviate from fundamentals, underpricing Unexpected impact of secondary market liquidity Lessons Complex products should not trade on simply trust Implicit support should be explicit Lack of market transparency creates false sense of safety Financial crisis is the ultimate test of financial innovation Song Han and Dan Li Liquidity Crisis, Runs, and Security Design 15