Leverage-Induced Fire Sales and Stock Market Crashes

Kelly Shue

Yale University and NBER

with Jiangze Bian, Zhiguo He, and Hao Zhou

1

• Excessive leverage and fire sales are believed to have been

major contributors to many past financial crises

• 1929 US stock market crash

• 2007/08 financial and housing crises

• 2015 Chinese stock market crash

• Theory of downward leverage spirals

• E.g. Brunnermeier and Pedersen (2009); Geanakoplos (2010)

• Tightened leverage constraints trigger fire sales, which then depress asset prices,

leading to even tighter leverage constraints

• General equilibrium theory featuring positive feedback loop

INTRODUCTION

2

• Limited empirical evidence on fire sales, and not in context of leverage with feedback loop

• Coval and Stafford (2007) and Edmans, Goldstein and Jiang (2012): fire sale of equities due to fund outflows

• Ellul, Jotikasthira, and Lundblad (2011): fire sale of downgraded corporate bonds due to regulatory constraints

• Campbell et al. (2011, foreclosed housing); Pulvino (1998, commercial aircraft)

• This paper: Direct evidence of leverage-induced fire sales

• Account-level trading data for margin accounts in Chinese stock market in 2015

• Examine role of shadow-financed margin trading and regulation

• For a study of the leverage amplification effect through the lens of a network contagion framework, see Bian, Da, Lou and Zhou (2017)

• Related to leverage and co-movement/liquidity: Kahraman and Tookes (2016 a,b)

INTRODUCTION

3

• Margin investors heavily sell their holdings when account-level leverage edges toward their maximum leverage limits

• Controlling for stock-date and account fixed effects

• Stocks that are disproportionately held by investors close to receiving margin calls experience high selling pressure and significant short run price declines that eventually reverse

• While regulated brokerage margin accounts owned a greater fraction of market assets, unregulated shadow margin accounts were the major drivers of leverage-induced fire sales

• Regulatory tightening announcements and price limits intensified fire sale pressure (these event studies also aid in identification)

PREVIEW OF RESULTS

4

BACKGROUNDChinese stock market crash

• Shanghai Composite Index: Started at around 3100 in Jan 2015, peaked at 5166 on

June 15, then collapsed to 3663 at the end of July

• Chinese stock market: 7.3 Trillion, second in size to US, 85% retail

Two types of margin accounts were popular starting in mid-2014

1. Brokerage-financed margin system

• Similar to US margin trading (initial margin, maintenance margin, etc.)

• Tightly regulated, with minimum initial margin and maintenance margin

2. Shadow-financed margin system

• “Mother account” (looks like a normal unlevered brokerage account with huge assets

and trading volume), linked through software to many levered “child accounts”

• Unregulated grey area: Lower maintenance margin, and larger cross-sectional

variation

5

BACKGROUND

The media and government allege that forced fire sale by leveraged accounts

(especially shadow accounts) were the leading cause of the crash

• May 22, 2015: CSRC (China Securities Regulation Commission) announced that brokerage firms should

“self-examine” shadow-financed margin accounts

• June 12 2015: CSRC released draft rules for a future ban on new shadow-financed margin accounts

• What do the data tell us?6

DATA

• Detailed account-level trading during the crisis (May-July 2015)

• Brokerage-financed margin accounts (Brokerage) from a leading brokerage firm,

cleaned sample represents ~5% of market share of brokerage margin service

• Shadow-financed margin accounts (Shadow) from a leading web-based peer-to-

peer lending platform

• Hard to estimate its market share: Best estimate for cleaned sample: ~5%

• Each individual account in both categories

• Daily stock holdings and trading

• Daily assets and debt, leverage = assets/(assets-debt)

• Account maximum allowable leverage (Pingcang Line, 平仓线)

• Stock market data: returns, volume, etc. 7

SUMMARY STATISTICS

8

LEVERAGE AND THE MARKET INDEX

• Leverage = Assets/Equity. • Asset-weighted and equity-weighted leverage are quite different!

index, right scale

leverage, left scale

9

ACCOUNT-LEVEL EVIDENCE

• ���� : Maximum leverage before the lender takes over

• So-called “Pingcang Line”

• Same for brokerage accounts, varies across accounts for shadow

• ����,� > ���� possible: Cannot sell if stocks hit +/-10% daily limit rule;

possible lender discretion in selling

• Proximity to the Pingcang Line:

• : Dummies for 10 equally-spaced bins by ��,�

,

,

1

1

j t

j tj

le vP

le v

10

,jk tI

ACCOUNT-LEVEL EVIDENCE

• Account-stock-date level regression:

•

• Stock-date fixed effect α�,� and account fixed effect α�

• Identification comes from account �’s time-varying proximity ��,�

• Robust to controls for account j’s recent past returns

• Leverage-induced selling implies that �� increases with �

10

, , , ,1

j j ji t k t i t j i tkk

I

,

Account 's net selling of stock at date

Account 's initial holding of stock at date ji t

j i t

j i t

11

ACCOUNT-LEVEL EVIDENCE

• Benchmark: classify accounts with � ≥ 6 as “fire sale accounts”

• Robust to using ��’s as weights to estimate fire sale exposure12

• Suppose proximity determines selling intensity

• Leverage still matters because it amplifies shocks

• For accounts with the same Proximity, those with higher leverage

should sell more aggressively: Their proximity will increase more for

a given drop in asset value

• Shadow sample: add leverage bins and interactions

10 5 5

, , , , , , ,1 1 11 0.6P L PL

k k kj Pj Lj Lj j ji t k t k t k t k t i t j i tk k k

I I I P

LEVERAGE AND PROXIMITY

13

LEVERAGE AND PROXIMITY

14

MARKET FEEDBACK

• Accounts with high proximity at the start of day t

should sell assets in both up and down market

conditions

• Need to sell to avoid margin call and/or deleverage

• Positive feedback of leverage spiral ⇒ stronger fire

sale effect in market downturn

15

MARKET FEEDBACK

16

STOCK-LEVEL EVIDENCE

• If stock � is disproportionately held by fire sale accounts, it should be

sold more heavily by these accounts

• Fire sale accounts: accounts with ��,� ≥ 0.6 at the beginning of �

• ����,� is stock �’s fire sale exposure

, , ,controls ji t i t i tFSE

,

Net selling of stock during date in fire-sale accounts

Outstanding shares of stock at date i t

i t

i t

,t

the beginnTotal ing shares of stock in fire-sale accounts at of date

Outstanding shares of stock at date i

i tFSE

i t

17

STOCK-LEVEL EVIDENCE

(1) (2) (3) (4)

Net selling of fire sale accounts

Fire Sale Exposure (FSE) 0.0996*** 0.102*** 0.102*** 0.102***

(0.0221) (0.0259) (0.0259) (0.0259)

Return Volatility X X

Size (Market Cap) X X

Turnover X X

Past 10-day cum. return X X

Past 10-day daily return X

Stock FE X X X

Date FE X X X

Observations 116,809 116,809 116,809 116,809

R-squared 0.144 0.186 0.186 0.187

18

NET SELLING BY FIRE SALE ACCOUNTS TO TOTAL VOLUME

19

• Sample restricted to stocks in the top decile of FSE on each day

• On average, net selling by fire sale accounts corresponds to 0.3% of volume

• Our sample = approximately 5% of margin market

RETURNS AND FIRE SALES

• We predict that stocks with high ��� underperform in the short-run

but not in the long-run

• Two methods

1. Double sort on past returns and ���; long-short strategy based on ���

2. Regression of stock returns from [t, t+X] on ��� with controls for volatility, market

cap, past returns, turnover, stock fixed effect, date fixed effect

• Which stocks fire sale accounts choose to sell is endogenous

• We use each stock’s fire sale exposure: fraction of shares held in fire sale accounts

20

PORTFOLIO RETURNS

• Double sort at the start of each day:

1. Sort stocks into quartiles by past returns ��,[����,���]

2. Sort each quartile into deciles by ����,�

• Long the top FSE decile and short the bottom FSE decile

• Leverage induced fire sales predict:

• Negative cumulative abnormal return, that reverts in long run

21

AVERAGE PORTFOLIO RETURNS

22

BROKERAGE VS SHADOW ACCOUNTS

23

FSE: BROKERAGE VS SHADOW

• Fire sale account cut-off � ≥ 0.6; higher leverage ≠ greater proximity

24

BROKERAGE VS SHADOW

25

BROKERAGE VS SHADOW : SELLING INTENSITY WHEN PROXIMITY EXCEEDS 1

26

SHADOW OR BROKERAGE?

• Regress CAR on FSE, constructed using just the Shadow or Brokerage samples• Coefficients represent the change in CAR for a std dev change in FSE• FSE constructed using the Shadow sample has larger effect and explanatory power

1 Day 3 Days 5 Days 10 Days 20 Days 40 Days

FSE of shadow -0.117*** -0.286*** -0.427*** -0.570*** -0.165* 0.0155

SE (0.0311) (0.0659) (0.0894) (0.0947) (0.0433) (0.844)

FSE of brokerage -0.0258*** -0.0883*** -0.0949** -0.0448 -0.0896*** 0.0300

SE (0.0107) (0.0236) (0.0341) (0.0391) (0.0222) (0.0337)

27

EVENT STUDY: REGULATION TIGHTENING

• Proposed regulations on shadow system released

• 5/22 (initial announcement) and 6/12 (detailed draft)

• Compare selling intensity in week before and after

announcements

• For both brokerage and shadow accounts

28

EVENT STUDY: REGULATION TIGHTENING

29

REGULATION: PRICE LIMITS

• Chinese stock market sets a daily price limit for

each stock: absolute return cannot exceed 10%

• Account-level selling intensity of each stock

should be stronger if other stocks cannot be sold

due to stock-specific price limits

• Fraction hitting limit = fractional value of account j’s assets

at the start of day t that consist of stocks that hit price limits

at some point on day t

30

REGULATION: PRICE LIMITS

31

• Because the fraction of holdings that hit price limits is correlated with returns,we control for the hypothetical portfolio return over day t assuming no trades

A DOUBLE SPIRAL?

32

• Loss spiral and margin/haircut spiral

• Figure from Brunnermeier and Pedersen (2008)

PINGCANG LINE: NEW SHADOW ACCOUNTS

33

• Pingcang lines never change within an account, but the average Pingcang line of new accounts varies positively with, and leads, the market index

• Ongoing: variation over time in interest rates?

Very few account openings

• Direct evidence of leverage-induced fire sales

• The closer to the maximum allowable leverage, the more investors sell (both

preemptive sales and forced sales)

• The resulting fire sale leads to negative abnormal returns in the short-run

• Feedback loop with market returns

• Regulated brokerage vs. unregulated shadow margin accounts

• Brokerage accounts dominate holdings, but had relatively low fire sale pressure

• Shadow accounts were the major force behind leverage-induced fire sales in

2015 stock market crash

• Regulation triggered and exacerbated fire sales in the short run

CONCLUSION

34

RETURNS FOLLOWING FIRE SALES

• Abnormal return is based on CAPM with stock

beta calculated using 2014 data

• ℎ = 1, 3, 5, 10, 20, and 40

• Model prediction

• �� < 0 for small ℎ but �� ≈ 0 for large ℎ

, , ,controlsi t h h i t i t hCAR FSE

35

RETURNS FOLLOWING FIRE SALES

• Standard errors clustered at date level• Controls: return volatility, market cap, past 10-day daily and cumulative

returns; turnover; stock fixed effect; date fixed effect

1 Day 3 Days 5 Days 10 Days 20 Days 40 Days

FSE -0.0978*** -0.259*** -0.357*** -0.413*** -0.180*** 0.0338

SE (0.0226) (0.0394) (0.0572) (0.0858) (0.0636) (0.0418)

36

LEVERAGE: BROKERAGE VS SHADOW

37

(Equity-weighted average – the conservative estimate)

ROBUSTNESS: CONSTRUCTING ��� BASED ON WEIGHTS

• Constructing stock level fire sale exposure ����,�

based on ��

• : number of shares of stock � in account �

• Numerator: weighted sum of shares of stock � in account �; if

account � belongs to group � then the weight is ��

• Again, leverage is measured at the beginning of date t

, ,

,tOutstanding shares of stock at date

j ji t k t kj

i

x IFSE

i t

,ji tx

38

PROXIMITY DISPERSION OVER TIME

39