Securitization and Insider Trading. Stephen G. Ryan New York University. Jennifer Wu Tucker University of Florida. Ying Zhou University of Connecticut

Transcription

1 THE ACCOUNTING REVIEW Vol. 91, No. 2 pp American Accounting Association DOI: /accr Securitization and Insider Trading Stephen G. Ryan New York University Jennifer Wu Tucker University of Florida Ying Zhou University of Connecticut ABSTRACT: Securitizations are complex and opaque transactions. We hypothesize that bank insiders trade on private information about banks : (1) securitization-related recourse risks, (2) not-yet-reported current-quarter securitization income, and (3) securitization-based business model sustainability. We provide evidence that proxies for each of these types of insider information are positively associated with insider trading. Specifically, we find that net insider sales in the 2001Q2 2007Q2 pre-financial crisis quarters predict not-yet-reported non-performing securitized loans and securitization income for those quarters, and that net insider sales during 2006Q4 predict writedowns of securitization-related assets during the 2007Q3 2008Q4 crisis period. We find that net insider sales are more negatively associated with banks subsequent stock returns in their securitization quarters than in other quarters. In supplemental analysis, we show that the above findings are driven by trades by banks CEOs and CFOs, and that insiders avoid larger stock price losses through 10b5-1 plan sales than through non-plan sales. Keywords: securitization; insider trading; opacity; banks. Data Availability: All data are available from public sources. I. INTRODUCTION We examine whether bank insiders exploit private information about an important type of complex structured-finance transaction, securitization, by trading for personal gain. Financial reporting requirements portray banks securitization-related risks in limited fashions, rendering these risks opaque to users of financial reports. We expect that bank insiders, particularly top executives, have informational advantages about these risks that enable profitable trading. In a typical securitization, the issuer (assumed to be a bank) transfers financial assets to a special-purpose entity (SPE), which sells asset-backed securities (ABS), i.e., claims to the future cash flows generated by the securitized assets, to outside investors. The SPE conveys the cash received from investors to the bank. Banks engage in securitizations for the economic purposes of transferring risks of the securitized assets and raising funds. Banks for which securitization is a business model, rather than one-off transactions, generally aim to earn income from securitizations and to use the cash received to originate securitizable assets on an ongoing basis. Banks also engage in securitizations for accounting purposes. While the applicable accounting rules have been tightened over time, banks continue to account for most securitizations as sales with the securitization SPEs unconsolidated. This accounting leaves the SPEs borrowings off the banks balance sheets and enables We thank Stephen Asare, Gauri Bhat (discussant), Stephen V. Brown, Michael Donohoe, Leslie D. Hodder (editor), Kathy Rupar, Stanley Veliotis, James Vincent, Dushyant Vyas, two anonymous reviewers, and participants at the 2011 American Accounting Association Annual Meeting and accounting workshops at the University of Florida, Fordham University, University of International Business and Economics, Shanghai University of Finance and Economics, and Yale University. Supplemental materials can be accessed by clicking the link in Appendix B. Editor s note: Accepted by Leslie D. Hodder. Submitted: May 2013 Accepted: July 2015 Published Online: July

2 650 Ryan, Tucker, and Zhou them to report gains on sale that front-load income, compared to the alternative of earning interest income on the securitized assets over time. We examine securitizations instead of other types of complex financial transactions (e.g., hedging) for three reasons. First, securitization is the most common type of structured-finance transaction. At the end of March 2007, the outstanding principal of ABS in the U.S. equaled $8.9 trillion, compared to $5.4 trillion of corporate bonds and $4.5 trillion of U.S. Treasuries (Cheng, Dhaliwal, and Neamtiu 2011). Although securitization volume for most asset classes other than agency mortgages fell dramatically during the financial crisis, volume has begun to rebound as the U.S. economy and financial markets recover. 1 Second, banks securitization-based business models worked well before, but poorly during the crisis, allowing us to examine large realizations of downside risks on banks existing exposures to these transactions. Third, banks public quarterly regulatory reports contain detailed and standardized data about their securitizations. Securitization is a contractually and economically complex transaction for which the issuing banks may be exposed to three types of risk. First, banks always retain some risk of providing recourse on previously securitized assets ( recourse risks ). Securitizations of credit-risky assets typically create risk-layered tranches of securities and other contractual asset and liability interests in the securitized assets. Issuing banks often retain risky and illiquid residual securities to credit enhance the senior securities sold to outside investors. Alternatively, banks may provide non-contractual ( implicit ) recourse. In essentially all securitizations, banks provide contractual representations and warranties that the securitized assets have the characteristics specified in the securitization prospectus. Banks violating representations and warranties are required to buy back the assets at par or other specified amounts if requested by the purchaser. Although historically viewed as distinct from recourse, buybacks of impaired securitized assets at prices above fair value due to actual or alleged violations effectively amount to recourse. Second, banks are exposed to uncertainty about their securitization income for the current period. This income depends on the prices that investors are willing to pay for the sold ABS and the estimated fair values of banks retained interests. Banks typically do not publicly disclose securitization income for a quarter until they file their financial and regulatory reports midway through the subsequent quarter. Third, banks for which securitization is a business model are exposed to uncertainty about the sustainability of this model, which requires continuing access to financing on acceptable terms that can evaporate quickly in credit crunches. Our study is motivated by and contributes to both the securitization and insider trading literatures. The former literature finds that securitizations accounted for as sales have attributes of both secured borrowings and sales (Niu and Richardson 2006; Landsman, Peasnell, and Shakespeare 2008; Chen, Liu, and Ryan 2008). Issuers time these securitizations to window-dress their balance sheets and manipulate reported earnings (Dechow and Shakespeare 2009; Dechow, Myers, and Shakespeare 2010). 2 Due to their complex economics and accounting, securitizations contribute to banks opacity (Cheng et al. 2011). We extend this literature, especially Cheng et al. (2011), by examining how bank insiders exploit their securitization-related information advantages through trading. The latter literature finds that insiders trade before significant price movements and the public disclosure of earnings (Lakonishok and Lee 2001; Ke, Huddart, and Petroni 2003). However, few studies in this literature examine specific sources of insiders information advantage. A notable exception is Aboody and Lev (2000), who identify research and development (R&D) as a source of private information and find higher insider trading profits at R&Dintensive firms than at other firms. By examining specific sources of insiders private information, researchers can identify and test for direct links between that information and insiders trading. Such tests help policymakers identify gaps in required disclosures and thereby limit insider trading. We extend this literature by examining securitizations as a source of bank insiders private information. To mitigate self-selection issues, our full sample includes bank holding companies only if they report securitized assets outstanding or non-zero securitization income in at least one quarter during our full-sample period of 2001Q2 2007Q2. We view securitization as a feasible choice for these Securitization Banks. We identify subsamples that correspond to the three types of securitization-related risks about which we expect insiders to have private information. The Recourse Risk subsample includes all bank-quarters with securitized assets outstanding at the end of the quarter; insiders in this subsample have private information about the banks recourse risks. The Securitization Income subsample includes all bank-quarters with non-zero securitization income reported for the current quarter or the previous quarter; insiders in this subsample have private information about the banks securitization income. The Crisis subsample includes all banks with non-zero securitized assets outstanding or securitization income in 2006Q4 on the verge of the financial crisis; 3 insiders in this subsample have private information about the (non)sustainability of the banks securitization-based business models. 1 In 2006, the issuance of mortgage-related ABS was $2.6 trillion (including $1.4 trillion of non-agency mortgages) and of other ABS was $268 billion. By 2010, the issuance of mortgage-related ABS had fallen to $2.0 trillion (including only $69 billion of non-agency mortgages) and of other ABS to $106 billion. In 2013, the issuance of mortgage-related ABS was $2.0 trillion (including $108 billion of non-agency mortgages) and of other ABS was $189 billion. See: 2 Barth and Taylor (2010) question whether Dechow et al. s (2010) findings are attributable to earnings management. 3 The financial crisis unfolded in waves, with the first wave (the subprime crisis) arriving with the announcement of significant losses on subprime mortgage-related positions by New Century Financial and HSBC Holdings on February 4, 2007 (Ryan 2008).

3 Securitization and Insider Trading 651 We develop three hypotheses about the association between bank insiders securitization-related private information and their trading that we test using the relevant subsamples and proxies for bank insiders private information about the specific risks involved. First, we hypothesize that bank insiders securitization-related private information in a quarter is contemporaneously positively associated with their trading volume in that quarter. We test this hypothesis using the Recourse Risk and Securitization Income subsamples. In the Recourse Risk subsample analysis, we proxy for insiders private information about recourse risks using quarter-end securitized assets, non-performing securitized loans, and retained securities, as well as charge-offs of securitized loans during the quarter. In the Securitization Income subsample analysis, we proxy for insiders private information about current-quarter securitization income using the absolute value of unexpected securitization income. As predicted, we find that these proxies are positively associated with bank insiders trading volume. This association is stronger for trades by banks CEOs and CFOs, who are more likely to possess private information, than by other insiders, and for the type of securitized assets most subject to implicit recourse, revolving consumer loans, than for other types of securitized assets. 4 Second, we hypothesize that net insider sales during a quarter predict unexpected values of specified securitization-related performance measures for that quarter or subsequent quarters that are reported after quarter-end. This hypothesis differs from the first in examining the direction of insider trades, as well as the predictive implications of these trades. We test this hypothesis using all three subsamples. In the Recourse Risk subsample analysis, we examine whether net insider sales predict unexpected non-performing securitized loans for the quarter. We find a significant positive association for trades by CEOs and CFOs, but not by other insiders. In the Securitization Income subsample analysis, we examine whether net insider sales predict unexpected securitization income for the quarter. We find a significantly negative association, again driven by CEO/CFO trades. In contrast, we do not find any association between net insider sales and unexpected non-securitization income. In the Crisis subsample analysis, we examine whether net insider sales in 2006Q4 predict write-downs of securitization-related assets during the financial crisis period of 2007Q3 2008Q4, our proxy for the breakdown of banks securitization-based business model during the crisis, 5 and find that this is the case. Hence, bank insiders appear to trade on specific types of securitizationrelated private information. In contrast, we do not find any relation between net insider sales and write-downs of nonsecuritization assets during the crisis. Third, we hypothesize that bank insiders profit from trading on securitization-related private information. Using the Recourse Risk and Securitization Income subsamples, we find economically and statistically significant negative abnormal stock returns for the three and six months after insider sales. For example, the median six-month stock return is 2.4 percent ( 3.9 percent) for the Recourse Risk (Securitization Income) subsample. Abnormal returns after insider purchases are consistently positive in both subsamples, but exhibit smaller magnitudes than after sales. The Recourse Risk and Securitization Income subsamples exhibit significantly more negative abnormal returns after insider sales than does a Control subsample that is comprised of Securitization Banks quarters without securitization activity. Furthermore, abnormal returns are significantly more negative after sales by CEOs and CFOs than by other insiders. Hence, bank insiders appear to avoid significant losses by selling shares before unfavorable securitization-related news becomes public. The Crisis subsample provides a vivid example of insiders selling early to avoid losses. Banks securitization-based business models became compromised as the performance of subprime mortgages and other types of credit-risky assets began to deteriorate before the financial crisis and the availability of financing evaporated early in the crisis. Due to banks central role in originating, holding, and securitizing these credit-risky assets, bank managers were among the first market participants to observe these adverse events. Crisis subsample banks experienced an average raw return of 64.8 percent during , 31.7 percent more negative than the market and 4.7 percent more negative than similarly sized banks. Net sales by bank insiders in the Crisis subsample totaled $1.19 billion in 2006Q4, over twice the amount in any other sample quarter; these sales enabled the insiders to avoid losses of $0.99 billion during the crisis. We conduct two supplemental analyses. First, we investigate the role of litigation risk in securitization-related insider sales. Beginning in October 2000, the Securities and Exchange Commission (SEC) allows insider sales under 10b5-1 plans that provide insiders with affirmative defenses against allegations of insider trading. Prior research finds that insiders avoid larger losses through 10b5-1 plan sales than through non-plan sales, consistent with these plans being used for private-information-based trades (Jagolinzer 2009; Shon and Veliotis 2013). We observe that 98.3 percent of plan sales for our full sample occur in securitization quarters when we expect bank insiders to possess securitization-related private 4 See Chen et al. (2008) for discussion of how implicit recourse applies only to certain types of securitized assets. 5 Ideally, we would use decreases in securitization volume and/or income (rather than write-downs of preexisting securitization-related exposures) during the financial crisis as proxies for the breakdown of banks securitization-related business model during the crisis. Such alternative proxies are difficult to develop and employ effectively, however, due to limitations of Y-9C report data (e.g., amounts for unaffected agency and highly affected nonagency mortgage securitizations are not disaggregated), the occurrence of non-trivial, apparent fire sale securitization volume during the crisis, and the attrition of the 33 Crisis subsample observations as the crisis unfolds.

4 652 Ryan, Tucker, and Zhou information. We find that the significant negative stock returns after insider sales in securitization quarters are driven by plan sales rather than by non-plan sales. Moreover, plan sales constitute 47 percent of sales by banks CEOs and CFOs, but only 23 percent of sales by other bank insiders, enabling CEOs and CFOs to avoid more losses than do other insiders. Hence, bank insiders appear to sell shares under 10b5-1 plans to provide cover for information-based trades, thereby mitigating litigation risk. Second, we test the claim that insider trading benefits investors by impounding insiders private information more quickly into stock prices (Manne 1970; Boudreaux 2009) using the future earnings response coefficient (FERC) framework (Collins, Kothari, Shanken, and Sloan 1994; Tucker and Zarowin 2006). We find that securitization and insider trading each separately reduce the informativeness of price with respect to future earnings and that insider trading does not alter the effect of securitization on this price informativeness. Hence, we find no evidence that securitization-related insider trading benefits investors. We believe our findings have implications for policymakers and investors. Our findings suggest that complex structuredfinance transactions such as securitizations require greater scrutiny from the SEC in enforcing insider trading rules and from investors in analyzing banks financial reports. In addition to their current approach of requiring firms to describe existing securitizations in detail, the SEC and the Financial Accounting Standards Board (FASB) should consider requiring management discussion and analysis (MD&A) or financial statement note disclosures that reveal bank managers information about the likelihood of banks future losses from providing recourse on securitizations and the sustainability of their securitization-based business models. The rest of the paper is organized as follows. Section II develops our hypotheses. Section III describes the sample and data. Section IV presents the research designs and test results. Section V provides supplemental analyses. Section VI concludes. II. HYPOTHESIS DEVELOPMENT Since the effective date of Statement of Financial Accounting Standards (SFAS) 125 in 1997, banks account for securitizations in which they cede control over the securitized assets as sales rather than as secured borrowings. 6 Although the FASB tightened the requirements for sale accounting with SFAS 140 and SFAS 166, in practice, banks continue to account for most securitizations as sales (Dechow et al. 2010). Under sale accounting, banks initially recognize retained interests at fair value. 7 For illiquid interests, banks must estimate fair value using internally developed or vendor models that banks do not (and likely cannot feasibly) fully describe in their financial reports. Holding a transaction constant, the valuation of retained interests directly and fully determines securitization income for that transaction: a dollar higher fair value assigned to a retained asset (liability) yields a dollar higher (lower) securitization gain on sale. For example, underestimation of default rates at the time of securitization increases the fair value of any retained junior securities and decreases the fair value of any recourse liability, thereby increasing the gain on sale. Effective in 2001, SFAS 140 requires banks to disclose considerable information about their securitizations, including: (1) the cash received and gains on sale recognized from securitizing assets by major asset type; (2) the fair value of each type of retained interest at the end of each reporting period; and (3) the sensitivity of these valuations to changes in key estimates. Effective in 2010, SFAS 166 and SFAS 167 expanded these required disclosures, especially about continuing involvements with securitized assets. Oz (2013) finds that these additional disclosures have improved the information available to investors. Despite these findings, significant aspects of securitization remain opaque to investors due to four limitations of securitization disclosures. First, banks describe securitized assets in aggregated and incomplete fashions in their financial reports and even in their securitization prospectuses. For example, securitization prospectuses often include statistics about the underwriting criteria (e.g., credit scores and loan-to-value ratios) that banks used in originating the securitized assets, but rarely provide information about risk-layering (e.g., the combination of low credit scores with high loan-to-value ratios). 8 Second, 6 For simplicity, we assume that banks do not consolidate securitization SPEs, as is usually the case, although the issuance of SFAS 167 made consolidation of certain types of SPEs (e.g., credit card master trusts and asset-backed commercial paper conduits) more common starting in Sale accounting with consolidation of securitization SPEs effectively yields secured-borrowing accounting for the consolidated entity. 7 The types of retained interests that generally accepted accounting principles (GAAP) require to be recognized at fair value at the time of sale have expanded over time. SFAS 125 and SFAS 140 required initial fair value measurement only for retained liability interests. SFAS 156 added servicing rights and SFAS 166 added all other asset interests. Before SFAS 166, GAAP required retained asset interests not required to be fair valued to be initially recognized at relative fair value-based allocations of the amortized cost basis of the securitized assets at the time of sale. 8 The information banks provide about risk-layering may have improved recently due to the SEC s revision of Regulation AB in August 2014, which requires extensive additional disclosures about securitization pools in securitization prospectuses and on an ongoing basis. See Acharya and Ryan (2016) for details about these requirements.

5 Securitization and Insider Trading 653 banks financial reports contain very little information about recourse risks associated with contractual representations and warranties and non-contractual implicit recourse (Niu and Richardson 2006; Landsman et al. 2008; Chen et al. 2008; Dou, Liu, Richardson, and Vyas 2014). Third, the volume and profitability of current-period securitizations are uncertain because they depend on (1) economic conditions, such as the receptivity of financial markets to securitization; (2) banks choices about whether to conduct and how to structure securitizations, given their financing needs and risk tolerance; and (3) banks exercise of discretion in fair-valuing retained interests and recording securitization income. Fourth, financial reports provide little information about the sustainability of banks securitization-based business models. Bank insiders likely have information advantages about all of these aspects of securitization, but the extant literature provides no empirical evidence about whether insiders exploit these advantages by trading for personal gain. Prior research finds that insiders exploit their information advantages by trading. For example, insider net purchases predict future stock returns (Lakonishok and Lee 2001), and insider sales predict breaks in strings of consecutive quarterly earnings increases three to nine quarters before the breaks occur (Ke et al. 2003). Assuming a firm s stock price impounds all information about the firm and its earnings measure the firm s overall operating performance, these studies are largely silent about the specific types of private information on which insiders trade. Identifying these types of information is important for two reasons. First, it helps researchers determine the control variables (e.g., risk factors or alternative sources of information) to include in empirical models to alleviate concerns that reported results are attributable to omitted correlated variables. Second, this identification clarifies the determinants of insider trading and, thus, the policy implications of the research. For example, Aboody and Lev (2000) identify R&D as a specific source of insider information and provide evidence that insiders trading profits are substantially larger at R&D-intensive firms than at other firms, consistent with R&D activities providing insiders with information advantages. Aboody and Lev (2000) conclude that requiring additional disclosures of firms ongoing and planned R&D activities would mitigate those advantages. We propose three hypotheses about the associations between bank insiders securitization-related private information and their trading. Each of these hypotheses reflects our expectation that insiders receive securitization-related information before its public release and that they exploit this private information by profitably trading. Although insiders private information is inherently unobservable, our study takes a step further than extant research by testing these hypotheses using subsamples and proxies intended to capture the private information that bank insiders possess about three types of securitization-related risks: (1) recourse risks, (2) uncertainty about current-period securitization income, and (3) uncertainty about securitization-based business model sustainability. We state all hypotheses as alternatives. Online Appendix A summarizes the hypotheses and indicates the tables that report the corresponding test results and supplemental analyses (the link to this and the other online appendices is provided in Appendix B). We first hypothesize a positive contemporaneous association between the amount of bank insiders securitization-related private information and their trading volume: H1: Bank insiders securitization-related private information is positively associated with their trading volume during the quarter. This hypothesis is similar to Cheng et al. s (2011) hypothesis of a positive contemporaneous association between a bank s securitization activity and measures of its information asymmetry. We next hypothesize that the direction and magnitude of insider trading during a quarter predict not-yet-reported securitization-related accounting performance measures (e.g., securitization income) for that quarter or subsequent quarters. Although we examine both insider sales and insider purchases, we expect the association to be stronger for insider sales because the major concerns raised by securitization pertain to downside risks: Will recourse be triggered? Will securitizations occur in the current quarter and be profitable? Is the securitization-based business model sustainable? Because most insider trades are sales, for simplicity, we state this and our third hypothesis in terms of net insider sales (insider sales minus insider purchases): H2: Net insider sales during a quarter are negatively associated with banks not-yet-reported securitization-related accounting performance for that quarter or subsequent quarters. Tests of this hypothesis can provide more direct evidence of whether insiders trade on valuable, private securitization-related information than do tests of our first hypothesis. Last, we hypothesize that insider sales ( purchases) are followed by stock price decreases (increases), more so in banks securitization quarters, when insiders likely possess more securitization-related private information, than in other quarters: H3: Net insider sales in banks quarters with securitization activity are more strongly negatively associated with the banks subsequent abnormal stock returns than are net insider sales in the banks other quarters.

6 654 Ryan, Tucker, and Zhou TABLE 1 Sample Selection Panel A: Full Sample (All Securitization Bank-Quarters) Bank-quarters with Y-9C filings during 2001Q2 2007Q2 and PERMCO 11,513 Exclude banks reporting zero securitized assets and securitization income each quarter of the sample period (8,402) Exclude bank-quarters with missing CUSIP (472) Exclude bank-quarters with missing insider trading data (639) Full sample 2,000 Panel B: Recourse Risk Subsample (with Risk of Providing Recourse on Securitized Assets) Full sample 2,000 Exclude bank-quarters with zero securitized assets at the end of the quarter (873) Recourse Risk subsample 1,127 Panel C: Securitization Income Subsample (with Uncertainty about Current-Quarter Securitization Income) Full sample 2,000 Exclude bank-quarters with missing securitization income (19) Exclude bank-quarters with zero securitization income in both the current quarter (t) and the previous quarter (t 1) (1,446) Securitization Income subsample 535 Panel D: Crisis Subsample (with Uncertainty about Securitization-Related Write-Downs) Full sample 2,000 Cross-section of full sample banks in fourth quarter of Exclude banks with zero securitized assets and securitization income (38) Exclude banks acquired in the first two quarters of 2007 (1) Crisis subsample 33 The Control subsample includes the 816 bank-quarters of the full sample that do not belong to any of the Recourse Risk, Securitization Income, or Crisis subsamples. III. DATA Although securitization occurs in many industries, as in most prior research, our sample includes only bank holding companies ( banks ). Our sample period begins in 2001Q2 when the Federal Reserve Board included Schedule HC-S, Servicing, Securitization and Asset Sale Activities, in banks quarterly regulatory Y-9C reports. Banks with more than $150 million ($500 million) in total assets before (after) March 2006 must file these reports. Schedule HC-S requires banks to report detailed and standardized data about their securitizations accounted for as sales in which they retain servicing rights or provide credit enhancement. As in Cheng et al. (2011), we end the sample period in 2007Q2 for the Recourse Risk and Securitization Income subsample analyses that examine insider trading during periods when banks securitization-based business models appeared to perform well. We extend the sample period for our Crisis subsample analysis of the breakdown of those business models during the financial crisis. Panel A of Table 1 summarizes the sample selection for the full sample. We require bank-quarter observations to have available Y-9C filings and non-missing PERMCO in the Federal Reserve Bank of New York s file that matches Y-9C filings to CRSP, 9 yielding 11,513 initial observations. To mitigate self-selection, we restrict the full sample to banks with securitized assets outstanding or non-zero securitization income in at least one quarter of our sample period. Although this restriction excludes 8,402 observations of banks for which securitization does not appear to be a feasible choice, the remaining 3,111 observations for Securitization Banks constitute 85.8 percent of the market capitalization of all banks, on 9 See:

7 Securitization and Insider Trading 655 FIGURE 1 Aggregate Securitized Assets and Securitization Income for the Full Sample Each Quarter This figure depicts aggregate securitized assets and aggregation securitization income for the full sample of 2,000 bank-quarters during each quarter of the 2001Q2 2007Q2 sample period. Panel A of Table 1 reports the construction of the full sample. average, during our sample period. We exclude 472 observations with missing CUSIP (Committee on Uniform Security Identification Procedures), which is necessary to merge Y-9C data with insider trading data from Thomson Reuters Insider Filing Data Feed. Following Seyhun (1992), Rozeff and Zaman (1998), Piotroski and Roulstone (2005), andcheng and Lo (2006), we limit insider trades to open market purchases and sales, which are most likely to reflect insiders private information. 10 The Thomson Reuters database does not contain data for 639 bank-quarters; it is unclear whether this is attributable to absence of insider trading or database incompleteness. We exclude these observations following Lakonishok and Lee (2001), Frankel and Li (2004), andpiotroski and Roulstone (2005). The resulting full sample includes 2,000 bankquarters for 130 unique banks, with the number of banks in a quarter varying from a maximum of 98 in 2004Q1 to a minimum of 67 in 2006Q3 and 2007Q1 2 (untabulated). Figure 1 depicts aggregate securitized assets outstanding and securitization income for the full sample for each quarter of the 2001Q2 2007Q2 sample period. Aggregate securitized assets generally increase over this period, peaking at $1.9 trillion in 2006Q4. Aggregate securitization income fluctuates considerably over time, with the maximum being $5.5 billion in 2004Q3. Panel A of Table 2 reports summary statistics for insider sales and purchases calculated as the number of shares traded multiplied by the trade price. Consistent with prior research, insider sales are a large multiple of insider purchases (about 13 [33] based on the mean [median] of the variables). Figure 2 depicts the total dollar amount of insider sales and purchases for the full sample during each quarter of the sample period. Insider trading varies moderately over time except for 2006Q4, when the amount is about 2.5 times higher than in any other quarter, indicating unusually active insider trading on the verge of the financial crisis. We identify three subsamples of the full sample in which bank insiders likely possess specific types of securitizationrelated private information. Panels B D of Table 1 summarize the selection processes for these securitization-treatment subsamples. The Recourse Risk subsample includes the 1,127 bank-quarters reporting a positive balance of securitized assets at the end of the quarter, with the number of banks in a quarter varying from a maximum of 58 in 2001Q3 to a minimum of 33 in 2007Q1 (untabulated). Because Schedule HC-S includes only securitizations accounted for as sales in which banks retain servicing rights or provide credit enhancement, banks reporting securitized assets generally are exposed to recourse risks. The Securitization Income subsample includes the 535 bank-quarters with non-zero securitization income in either the current quarter or the previous quarter, with the number of banks in a quarter varying from a maximum of 29 in 2001Q4 to a minimum of 15 in 2006Q3 (untabulated). Of the Securitization Income subsample 10 Most other insider trades are between the insiders and their firms. These trades tend to be driven by stock option grants and other stock-based forms of compensation.

8 656 Ryan, Tucker, and Zhou TABLE 2 Descriptive Statistics Panel A: Insider Trading Summary Statistics for the Full Sample n Mean Median Std. Dev. 25% 75% Insider sales ($000) 2,000 3, , ,743 Insider purchases ($000) 2, , Panel B: Variable Means and Medians for the Subsamples Recourse Risk Subsample Securitization Income Subsample Crisis Subsample n Mean Median n Mean Median n Mean Median NetTrade 1, SB 1, SBM 1, SBCON 1, SBCOM 1, NPSL 1, Chgoff_Sec 1, Retained 1, jusij MVE ($million) 1,127 20,610 4, ,982 12,378 Turnover 1, MB 1, jdroaj 1, jpastretj 1, jfutureretj 1, StdRet 1, StockComp 1, ExecOwnership 1, NetSale 1, Sale 1, Buy 1, NetSale_Exec 1, Sale_Exec 1, Buy_Exec 1, NetSale_Other 1, Sale_Other 1, Buy_Other 1, Writedown_S 07Q3 08Q Writedown_NS 07Q3 08Q NetSale 06Q Sale 06Q Buy 06Q Log(TA 06Q4 ) Loan 06Q Return t,tþ3 1, Return t,tþ6 1, Return crisis Total 1, (continued on next page)

9 Securitization and Insider Trading 657 TABLE 2 (continued) Panel C: Pairwise Correlations of Main Variables for the Full Sample NetTrade SB NPSL Chgoff_Sec Retained jusij log(mve) log(turnover) MB jdroaj jpastretj jfutureretj StdRet StockComp ExecOwnership Table 1 reports the construction of the full sample and subsamples. In Panel A, insider sales and insider purchases for a bank in a quarter are calculated as the number of shares traded multiplied by the trade price, summed across all insiders of the bank in the quarter. In Panel C, Pearson (Spearman) correlations are reported above (below) the diagonal. Correlations that are statistically significant at the 1 percent level are in bold. The means and medians of Return t,tþ3, Return t,tþ6, and Return crisis reported in Panel B are for the bank-quarter variables used in the bank-quarter-level regression analysis reported in Online Appendix B; analogously defined trade-level Return t,tþ3 and Return t,tþ6 are used in the trade-level analysis reported in Table 7. Appendix A contains the variable definitions. observations, 89.3 percent also belong to the Recourse Risk subsample. 11 The Crisis subsample includes the 33 banks reporting either a positive balance of securitized assets or non-zero securitization income in the fourth quarter of 2006 on the verge of the financial crisis; all of these observations appear in one or both of the Recourse Risk and Securitization Income subsamples. We refer to the bank-quarter observations in the full sample that do not belong to either the Recourse Risk or Securitization Income subsamples as the Control subsample. We contrast the Recourse Risk and Securitization Income subsamples to the Control subsample in the analyses of the profitability of insider trading. We do not contrast the Crisis subsample to the Control subsample, however, because we do not expect the time-distributed Control sample observations to be comparable to banks with securitization exposures in a single quarter on the verge of the financial crisis. Tests of H1 IV. PRIMARY ANALYSES H1 predicts that the amount of bank insiders securitization-related private information is positively associated with insider trading volume during the quarter. We test this hypothesis by estimating this equation using the Recourse Risk and Securitization Income subsamples: NetTrade t ¼ a 0 þ a 1 Securitization t þ a 2 logðmve t Þþa 3 logðturnover t Þþa 4 MB t þ a 5 jdroa t jþa 6 jpastret t j þ a 7 jfutureret t jþa 8 StdRet t þ a 9 StockComp t þ a 10 ExecOwnership t þ a 11 StockComp t ExecOwnership t þ e t : ð1þ 11 Securitization income in a quarter may be zero for banks with non-zero securitized assets in the quarter, and vice versa, for various reasons. For example, some banks have positive securitized assets in a quarter from securitizations in prior quarters, but conduct no securitizations and, thus, earn no securitization income in the current quarter. Some banks record zero securitization income despite conducting securitizations in a quarter. Some banks conduct securitizations during a quarter for which they do not retain servicing rights or provide credit enhancement and so do not disclose the securitized assets on Schedule HC-S.

10 658 Ryan, Tucker, and Zhou FIGURE 2 Aggregate Insider Sales and Purchases for the Full Sample Each Quarter This figure depicts the aggregate dollar amounts of insider sales and purchases for the full sample of 2,000 bank-quarters during each quarter of the 2001Q2 2007Q2 sample period. Panel A of Table 1 reports the construction of the full sample. Following Piotroski and Roulstone (2004), the dependent variable is NetTrade, the absolute dollar amount of sales minus purchases by all of the bank s insiders during the quarter, multiplied by 100 for presentation purposes, and divided by beginning-of-quarter market value of equity. 12 NetTrade is unaffected by equal increments to insider sales and purchases for a bank in a quarter and, thus, treats such increments as having perfectly offsetting implications. The primary explanatory variable, Securitization, stands in for our proxies for bank insiders securitization-related private information, which differ across the subsample analyses. In the Recourse Risk subsample analysis, we proxy for this information using four variables identified by Cheng et al. (2011): (1) quarter-end securitized assets, SB, which captures the size of the exposure to recourse risks; (2) quarter-end non-performing securitized loans, NPSL, and (3) net charge-offs of securitized loans during the quarter, Chgoff_Sec, which capture the recent performance of the securitized assets; and (4) quarter-end retained securities from securitizations, Retained, which captures the extent of the primary form of credit enhancement. We expect banks recourse risks to increase with each of these proxies. Bank insiders private information involves knowing the values of these variables before their public disclosure, as well as the implications of the variables for the likelihood that the bank provides recourse on the securitized assets and the amount of losses resulting from that provision. In the Securitization Income subsample analysis, to develop the proxy for the amount of bank insiders securitizationrelated private information, we first estimate this equation predicting securitization income, SI: SI t ¼ a 0 þ a 1 SI t 1 þ a 2 SB t 1 þ a 3 Q1 t þ a 4 Q2 t þ a 5 Q3 t þ year fixed effects þ e t : On the right-hand side of Equation (2), we include securitization income for the previous quarter, which we expect to be the best predictor of current-quarter securitization income. Following Dechow et al. (2010), we divide securitization income by beginningof-quarter book value of equity. We include beginning-of-quarter securitized assets to capture the strong tendency for banks that have previously conducted securitizations to continue to do so. Finally, we include fiscal quarter dummies to control for seasonality, and year fixed effects to control for macroeconomic and financial market factors. We measure unexpected securitization income, USI, as the estimated residual in Equation (2) and use the absolute value of unexpected securitization income, jusij, as the proxy for ð2þ 12 Piotroski and Roulstone (2004) scale their measure by trading volume. We scale instead by beginning-of-quarter market value of equity, because trading volume trends upward strongly over time with the growth in high-frequency trading after their sample period ends in 2000.

11 Securitization and Insider Trading 659 the amount of bank insiders securitization-related private information. Because this proxy plays a more central role in the test of H2 than in the test of H1, we discuss the estimation of Equation (2) in the section devoted to the test of H2. Following Piotroski and Roulstone (2004) and Cheng and Lo (2006), we control for the following variables in Equation (1). We control for the natural logarithm of banks beginning-of-quarter market value of equity, log(mve), which we expect to be negatively associated with insider trading because larger firms tend to have stronger corporate governance and higher media scrutiny. We control for banks stock liquidity using the natural logarithm of trading volume divided by beginning-of-quarter number of shares outstanding, log(turnover), because liquidity affects insiders ability to trade without moving the price. We control for beginning-of-quarter market-to-book ratio, MB, and for the absolute value of the change in quarterly net income from the previous quarter to the current quarter divided by beginning-of-quarter total assets, jdroaj, because we expect insiders of banks with higher growth and more variable profitability to have greater information advantages. We control for the absolute value of the bank s stock return in the previous quarter minus the index return of banks of similar size, jpastretj, because prior research finds that insiders are more likely to trade after larger price movements. We also control for the absolute value of the bank s return minus the average return of banks of similar size in the 12 months following the current quarter, jfutureretj, to capture insiders private information about banks future performance unrelated to securitization (Ke et al. 2003). We control for the standard deviation of daily stock returns in the 12 months before the current quarter begins, StdRet, because managers may trade to reduce their exposure to bank risks other than recourse risks. We control for the fraction of the bank s shares owned by its reportable executives at the end of the most recent fiscal year, ExecOwnership, and for the dollar value of restricted stock and option grants divided by total compensation in the most recent fiscal year averaged across these executives, StockComp, because managers with high stock ownership or recent stock-based compensation tend to trade to reduce this exposure (Cziraki 2015; Ofek and Yermack 2000). We further include the interaction of ExecOwnership and StockComp to capture the effect of the combination of stock ownership and stock-based compensation on insiders tendency to trade. We are able to calculate ExecOwnership and StockComp using data from Execucomp for 995 observations (about 50 percent of the full sample); we hand-collect data for the remaining observations from the banks proxy statements. Panel B of Table 2 reports descriptive statistics for the variables in Equation (1). In the Recourse Risk subsample, securitized assets (undeflated SB), on average, equal a sizeable 21.5 percent of total assets. The other recourse risk proxies have considerably smaller values. In the Securitization Income subsample, securitization income (undeflated SI) equals 34.3 percent of net income, on average (untabulated). Many of the variables have skewed distributions, as evidenced by the large differences between their means and medians. Panel C of Table 2 reports pairwise Pearson and Spearman correlations for the full sample. Because of data skewness, we discuss only the Spearman correlations. NetTrade is negatively and insignificantly correlated with SB and jusij, respectively, apparently inconsistent with H1. These correlations are likely driven by bank size: log(mve) is highly negatively correlated with NetTrade, but highly positively correlated with SB and jusij. Panel A of Table 3 reports the estimation of Equation (1) using robust regression to mitigate the effects of the data skewness. Robust regression iteratively reweights observations until the estimated coefficients converge. This method is superior to traditional methods of dealing with outliers, such as winsorization and truncation, because it identifies and assigns weights to outliers in a multivariate distribution (Anderson 2008; Leone, Minutti-Meza, and Wasley 2012). The first four columns of the panel use the Recourse Risk subsample with the four proxies for bank insiders securitization-related private information about recourse risks included one at a time. Consistent with H1, the coefficients on SB, NPSL, Chgoff_Sec, and Retained are significantly positive at (t ¼ 2.18), (t ¼ 2.84), (t ¼ 4.44), and (t ¼ 3.15), respectively. The fifth column uses the Securitization Income subsample with jusij as the proxy for uncertainty about current-quarter securitization income. Again consistent with H1, the coefficient on jusij is significantly positive at (t ¼ 14.43). The control variables log(mve), log(turnover), and MB have consistently significant coefficients of the predicted signs across the models, except for the positive, but insignificant, coefficient on MB in the Securitization Income subsample analysis. The coefficients on the other control variables are less consistently significant with the predicted sign. 13 Overall, the results reported in Panel A of Table 3 indicate that insiders trade more when they possess better securitization-related private information. We refine the above analyses in two ways to probe our interpretation of the positive estimated coefficients on the Securitization proxies in Equation (1) as attributable to bank insiders trading on securitization-related private information. First, we decompose insider trades into those by CEOs and CFOs versus by other insiders, because prior research provides evidence that CEOs and CFOs possess better private information than other insiders. For example, Cheng and Lo (2006) find that the number of bad-news earnings forecasts is more strongly positively associated with planned purchases by CEOs than by other 13 We expect the coefficient on jdroaj to be positive. Results in the Recourse Risk subsample analysis are largely consistent with this expectation, but this coefficient is insignificant in the Securitization Income subsample analysis. The coefficient on jpastretj is weakly significantly negative in most of the models. We expect the coefficient on jfutureretj to be positive and find a weakly positive coefficient in the Securitization Income subsample analysis, but a negative coefficient in the Recourse Risk subsample analysis. The coefficient on StdRet is insignificant. The coefficient on ExecOwnership is significantly positive, as expected, in the Recourse Risk subsample analysis, but not in the Securitization Income subsample analysis. The interaction term of StockComp and ExecOwnership is significantly positive only in the Securitization Income subsample analysis.