Can short sellers inform bank supervision?

Following Berger et al. (2000), timeliness is the marginal ability of short sellers signal to predict supervisory ratings and vice versa. Accuracy measures the marginal value of short sellers or supervisors predictive ability about bank future performance. We thank the anonymous referee for the suggestion to set up the research question on timeliness and accuracy dimensions.

“CAMELS” is an acronym denoting capital adequacy (C), asset quality (A), management (M), earnings (E), liquidity (L), and sensitivity to market risk (S).

We thank the anonymous referee for making this suggestion to examine heterogeneity.

http://​www.​occ.​gov/​topics/​laws-regulations/​enforcement-actions/​enforcement-actions-types.​html

Short interest is the total number of shares of a particular stock that investors have sold short but have not yet been covered or closed out. The short interest ratio (SIR) is the ratio of the number of uncovered short positions to the total number of shares outstanding. The SIR is the standard measure of short interest. The empirical research on short selling such as Asquith et al. (2005) and Boehmer et al. (2010), among others, use the SIR. A higher SIR in a stock discloses a higher level of negative sentiments about that firm. Information on short interest is published twice a month. The SIR data are available every fortnight since January, 2007 and every month prior to January, 2007.

CAMELS changes usually coincide with on-site examination, which typically occur every 12 to 18 months. New financial statements are published quarterly. We examine quarterly changes to balance out the infrequent changes in the CAMELS and the more frequent changes in the SIR and other stock market variables following Berger et al. (2000).

In unreported results, we also consider charge-off ratio, Tier 1 risk-based capital ratio, and loan growth; the results point to similar conclusions.

Aiken et al. (1998) show that investors impound adverse information within 15 min or 20 trades in the Australian Stock Exchange that identifies short sales publicly. In abusive naked short selling, stocks are short sold, without locating or borrowing stocks. In a manipulative strategy known as short and distort strategy, stocks are sold short and then short sellers spread a false rumor to drive down the stock price. The US Securities and Exchange Commission (SEC) are likely to prosecute abusive and manipulative short selling.

Contrarian investors argue a higher short interest is good news because the buying pressure increases when short sellers cover their open positions.

We base our definition of very large banks on the Dodd-Frank Act (2010) threshold: Section 165(i) (2) of the Dodd-Frank Wall Street Reform and Consumer Protection Act (Public law 111–203, 124 Stat. 1376, July 2010).

More details on these bans can be found at http://​www.​sec.​gov/​news/​press/​2008/​2008-143.​htm and https://​www.​sec.​gov/​news/​press/​2008/​2008-211.​htm

We start from the first quarter of 2004. Because we use the changes in variables, we could not include the first quarter of 2004 in our regression analysis.

http://​www.​optionsclearing.​com/

Our results do not change qualitatively when we use different measures of CAMELS ratings, such as, the median or mean CAMELS rating of the BHC subsidiaries as well as the asset-weighted CAMELS rating of subsidiary banks. We exclude BHCs where national banks (for which we have CAMELS data) comprise less than 50 % of the bank’s assets. This is to ensure that our CAMELS ratings are representative of the BHC. Though we have taken care to include the largest subsidiary, we examine whether a significant change in part of the BHC affects the short interest of the entire holding company, for which the stock is listed. Therefore, our tests are inherently designed not to find a significant effect of CAMELS changes in short interest.

For robustness, in unreported results, we also include a control denoting whether the inspection occurred in a given quarter. We find no evidence that recent inspections are linked with subsequent changes in short interest.

For logit specifications considering only upgrades or downgrades, we replace the time dummies with a set of macroeconomic controls including fed funds rate, quarterly change in the fed funds rate, unemployment rate, quarterly change in unemployment rate, volatility index (VXO), and mean regulatory intensity (i.e., average CAMELS ratings for all banks during the quarter). We do so because there are not enough of these events in many quarters causing the logistic procedure to drop observations or fail when we include time dummies. We also use macroeconomic controls in place of time dummies in sub-sample tests (Table 8) for the similar reasons. In additional unreported tests, we use a linear probability model with the time dummies included and achieve similar results.

In unreported robustness tests, we also estimate model (2) with an OLS and find similar results.

In additional unreported tests, we estimate specifications analogous to column (4) and column (8) but respectively exclude the lags for CAMELS changes and SIR changes. These tests show a lack of an increase in the adjusted-R-square in models where the CAMELS lags are included but an increase in the pseudo-R-square and the Chi-square goodness of fit statistic when the SIR lags are included.

This calculation is based on the fact that the reported coefficients in Table 3 for the ordered logit regressions are the logs of the odds ratios and the standard deviation of SIR change is about 0.0131; therefore we estimate 0.0131x19.232 = 0.2512 to be the increase in the log odds of CAMELS change for one standard deviation increase in SIR change.

There are only 12 significant upgrades in the sample compared to 36 significant downgrades.

In unreported results, we also consider the charge-off ratio, Tier 1 risk-based capital ratio, and the loan growth; the results point to similar conclusions.

In unreported tests, we replace the Δ SIR with separate variables for an increase and a decrease in SIR and find results similar to that in Table 6. The results, which can be obtained from the authors upon request, generally confirm the results of Table 6 but also suggest some asymmetry in the directional impact of changing SIR on performance.

There is a weaker and insignificant coefficient in column 7 of Table 7 compared to the result in column 8 of Table 3 that indicates that SIR changes Granger-cause a CAMELS change only for banks with high bid-ask spreads. We confirm this result in additional unreported tests.

As an additional robustness check, we examine whether bank mergers drive our results. Mitchell et al. (2004) show that merger arbitrage short selling increases the price pressure on acquirers. In our sample 8.42 % of the BHCs have at least one subsidiary involved in a significant merger. The FDIC defines a significant merger as one where assets change by 25 % or more. We exclude BHCs with such merger activity from the sample and examine whether the change in SIR is still significant. The results, which we not report to conserve space, show that merger activity does not substantially affect the timeliness and accuracy of short sellers’ signals. We thank the anonymous referee for the suggestion.

Engelberg et al. (2012) use the exempt and non-exempt marking available in the Reg SHO data to show that client trades are more informative than market maker trades. We do not have such a distinguishing feature in the short interest data.