Banks, Bears, and the Financial Crisis

Perhaps history’s first ban on short selling was imposed in Amsterdam in 1610 (Bris et al. 2007).

See, for example, Brunnermeier and Pedersen (2005) and Goldstein and Guembel (2008).

The Panic of 1907, for example, is believed to have originated in a failed short squeeze that raised doubts about the solvency of the banks and brokerages that had enabled the scheme. See Bruner and Carr (2007).

See, for example, evidence on contagion across asset classes due to institutional trading in Boyer et al. (2006) and Manconi et al. (2012).

See Berger et al. (2008) for pre-crisis evidence on U.S. bank capital ratio management and Minton et al. (2009) for evidence on their use of credit derivatives.

Until the Reg SHO and NYSE short trade data became available, almost all research relied on proxies for shorting such as borrowing costs, short interest, or institutional ownership. For example, a notable contribution to this literature, Jones and Lamont (2002), is based on hand-collected stock borrowing costs from newspapers from December 1919 to October 1933, spanning the October 1929 crash period. Another good example is Liu et al. (2010), who use monthly NYSE short interest to study shorting around recent mortgage write-down events.

We examined shorting for ten NYSE-listed ETFs and closed end funds that hold financial stocks and have shorting activity. Summary statistics (available on request) indicate daily aggregate shorting of these securities declined during the crisis. Beyond the markets studied by Battalio and Schultz (2011) and Ni and Pan (2011), other derivatives markets may have provided means for shorting financial stocks indirectly. However, a good example, S&P Financial index futures traded on the CME, has never enjoyed significant trading volume.

They also report (Table V) that, during the period they study (January 2000 to April 2004), most shorting is conducted by institutional investors. That most comes from hedge funds (Goldman Sachs 2010) further fuels the controversy about short selling.

See, for example, the model of De Long et al. (1989) in which noise trading reduces investment and consumption.

Such predictions can be complex. For example, on page 577, Diether et al. (2009) state “In a market with wide dispersion in reservations values, limit orders posted by (nonstrategic) competing liquidity providers result in narrower spreads. As opinions converge, volatility should fall and spreads should widen.”

This data is available for purchase from NYSE/Euronext and is distinct from the Regulation SHO database which covers the period from January 2005 to 6th July 2007.

Stock and option market makers were allowed to short for certain purposes, so shorting did not vanish entirely. Further sub periods are possible given events such as the suspension of the uptick rule, changes in rules for naked shorting, and expansion of the list of financial firms for which shorting was severely restricted. Early work on our paper included more sub periods but results did not differ any more substantially across periods than our current two-period study.

The SEC’s emergency order claimed to cover 799 stocks, but only 797 were actually listed in the order.

We also would have excluded closed-end funds, ETFs, and real estate investment trusts, but there were none among the 204 NYSE firms on the SEC’s list.

Given that most financial institutions were placed on the short restriction list, this implies that matched firms cannot be drawn from the same industries as the sample firms.

The table also indicates the control sample includes 11 financial firms that were apparently not distressed and not placed on the shorting ban list in September 2008. Other partitions of the data could define the sample as all financial firms and construct the control sample from non financial firms. However, this is unlikely to affect our results which, as we discuss below, show little difference between the financial and control samples.

Boehmer et al. (2008) report many findings that are robust to whether shorting is measured with number of short trades, number of shares shorted, or shorting as a proportion of volume. They indicate that shorting scaled by volume minimizes correlation with returns induced by size, book-to-market, and volume effects.

For each trade, a trade price below (above) the midpoint of bid-ask a price is classified as seller-initiated (buyer-initiated). If, the trade occurs at the bid-ask midpoint, it is classified as seller initiated (buyer–initiated) if the trade price is lower (higher) than the preceding trade price. The literature remains unsettled as to which classification method is best. For example, Ellis et al. (2000) find that the Lee and Ready (1991) method is slightly more likely correct (81.05 %) than the quote rule (76.4 %) or the tick rule (77.66) in classifying Nasdaq trades of 313 firms from September 1996 to September 1997, though all three are less successful at classifying trades away from quotes, which are more likely with large trades during very active markets. Chakrabarty et al. (2012) confirm that the algorithm works well over more recent INET trades. Finacane (2000) rates the Lee and Ready (1991) and tick tests about equal in classifying NYSE trades for 144 firms from November 1990 to January 1991, though the tick test is better at computing signed volume and effective spreads.

Many of the findings in Tables 5 and 6 are broadly consistent with Comerton-Forde et al. (2011): for the period January to August 2008 (part of our Crisis period), passive short traders appear to supply liquidity and trade against market trends.

Comparable plots for passive shorting and for shorting of matched firms are available on request.

An alternative approach is studying the “permanent price impact” (Linnainmaa and Saar 2012) over all individual short trades.

The formal structure of information available to market participants about shorting is as follows. NYSE has always released a monthly summary of short interest, while a daily summary (see www.​nyxdata.​com/​page/​875) only become available in late July 2009 after the end of our sample period. Informal knowledge of shorting activity and other aspects of order flow presumably varies across market participants.

Given the ability of traders to break large trades into small orders, it is not relevant to identify large short selling events based on trade size.

Note that our short sample period and emphasis on daily and intraday data precludes use of monthly sentiment indicators as developed in Baker and Wurgler (2006). For a related application to hedge funds, see Boyson et al. (2010).

John Hancock Bank and Thrift Opportunity Fund (BTO), Financial Trends Fund (DHFT), and First Trust Specialty Finance (FGB).

Knez et al. (1994) decompose money market yields and find that the 3-month eurodollar yield (and other credit-sensitive yields) displays substantial loading on an unobserved third factor which the 3-month T-bill displays only trivial loading on. Given that we cannot observe intraday Treasury bill yields to match the intraday Eurodollar yields implied in futures prices, the intraday Eurodollar yields may be thought of as a noisy proxy for the TED spread.

Given the 5-min and daily frequency of our tests, we cannot employ weekly liquidity indicators as in Table 5 of Longstaff (2010).

Recent research suggests that short selling can have less impact than ordinary selling. See Comerton-Forde et al. (2011) and Chakrabarty et al. (2012).

For sample firms and daily intervals during the Crisis period, for example, the average big shorting event is about 0.42 for aggressive shorting and 0.49 for passive shorting, while the average big aggressive liquidation event is about 68 %. Corresponding averages for matched firms are slightly smaller, 0.39, 0.44, and 0.67 respectively.

“…their influence (for both good and evil) is a little more than a drop in a bucket and something less than a hill of beans”, Schwed (1940), page 92.

“European short-selling ban comes under attack”, www.​ft.​com, 12th August 2011.

See http://​en.​wikipedia.​org/​wiki/​Naked_​short_​selling#Developments.​2C_​2007_​to_​the_​present.

Lacking transactions records tagged as naked short sales, some authors proxy for naked shorting with daily SEC records of failures-to-deliver (Fotak et al. 2009; Boulton and Braga-Alves 2010).

A recent working paper, Jones et al. (2012), studies the impact of mandatory disclosure of large short positions in three European countries but finds little impact except around rights issues.

See, for example, Ben-David et al. (2012) on hedge fund selling in 2007 and 2008.

See, for example, Griffin et al. (2011) on the tech stock boom and bust.