Variable Definitions

The total dollar value of accruals is calculated using the cash flow measure provided by Hribar and Collins (2002):

𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖𝑖𝑖 = 𝑁𝑁𝑁𝑁𝑖𝑖𝑖𝑖− 𝑂𝑂𝐴𝐴𝑁𝑁𝐶𝐶𝑂𝑂𝑖𝑖𝑖𝑖 (2)

where NI represents net income and OANCF represents net cash flow from operations. The total dollar value of accruals is scaled by the average total assets of the firm.

While Sloan is the seminal study and its prominence suggests that any trading strategy that exploits the accruals anomaly might be influenced by it to some degree. This paper deliberately uses Hribar and Collins’ (2002) definition for construction of accruals instead of Sloan’s (1996) for two reasons. Firstly, this paper seeks to avoid calculating accruals using an alternative definition to try and eliminate data-snooping biases that might arise. While numerous papers have focused on specific types of accruals or refined the definition of accruals provided by Sloan (1996), the returns from all of these strategies are explained by the recent crop of factor models (Hou et al (2015), Fama and French (2015) and Stambaugh and Yaun (2017)). This paper follows Hou et al. (2017), who use Hribar and Collins’ (2002) definition to measure accruals for the Sloan (1996) instead of Sloan’s (1996) own definition post-1988. They underscore that this switch is crucial as it minimizes measurement errors that can arise from non-operating activities like acquisitions. It should be noted that Fama and French (2015) used their own definition, which measured accruals as the change in operating working capital per split-adjusted share from year t-2 to year t-1 scaled by book equity per share in year t-1. While there is still a lack of clear consensus, this paper believes that it is apt to use the cash flow measure as defined by Hribar and

31

Collins (2002) as it has been documented to reduce measurement error in comparison to the balance sheet measure contributed by Sloan (1996). Secondly, Green et al. (2011) note that hedge funds apply far more sophisticated approaches than reported in the academic literature. Additionally, DeFond and Hung (2003) show that analysts’ cash flow forecasts they are the result of sophisticated models that predict accruals instead of just simple adjustments of the earnings forecasts for routine items like depreciation and tax. Thus, it is a reasonable to assume that hedge funds that trade on the accruals anomaly probably employ a more sophisticated methodology too. So, any influence that Sloan (1996) had on hedge strategies has probably disappeared as they came up with more refined definitions. Additionally, it is assumed that these hedge funds use a cash flow measure instead of a balance sheet measure as it has been widely documented to reduce the measurement error generated by non-operating activities. While under ideal circumstances this paper would have liked to measure accruals using the same

methodology as practitioners, these naïve assumptions are necessary as this paper has no way of identifying and replicating those more refined approaches. It merely serves as a preliminary investigation into the magnitude of returns earned by the accruals anomaly in the sample of firms that have weak internal controls.

Sloan (1996) constructed portfolios once a year on June 1 of year t using only firms with a December 31 fiscal year end. Green et al. (2011) highlight that the accruals anomaly has attracted the attention of various large investment entities, who have deployed a considerable amount of AUM to exploit it. They state that sophisticated investors are unlikely to ignore quarterly data. The fact that most sophisticated investors go to great lengths to uncover

32

support to this claim.23 Green et al. (2011) construct an applied approach, in which they rebalance long/short positions monthly rather than annually. They construct hedge portfolio using a value-weighted basis. They only use the largest 3,000 firms to construct the portfolios.

This paper follows Hou et al. (2017) methodology for replicating the accruals anomaly. Previous studies have imposed restrictions on the minimum cutoff for the share price of a firm like $1 or $3 and have dropped firms that don’t meet it. Following, Hou et al. (2017), this paper doesn’t employ any such minimum cutoff for the share price of the firm for inclusion in the sample. It also doesn’t restrict the sample for portfolio construction to firms that meet a certain size cutoff. This is done to maximize the sample size as Lev and Nissim (2006) document that extreme accruals firms have low share prices and are relatively smaller. Additionally, Table 2 shows that firms with weak internal controls also possess these characteristics. Thus, this paper’s sample contains a larger number of firms with weak internal controls and extreme accruals firms. Since this paper is a preliminary investigation into whether or not the accruals anomaly can be exploited in firms with weak internal controls, harsh restrictions are not imposed initially. Additionally, these restrictions weren’t imposed by Hou et al. (2017), who replicate 447 anomalies, either. Following Hou et al. and prior literature on the accruals anomaly, financial firms are excluded as the same definition cannot be used to calculate accruals for them. The monthly returns obtained from CRSP are adjusted for delisting wherever it is necessary. Following Green et al. (2011), delisting returns are set to -35% for NYSE/AMEX firms and - 55% for NASDAQ firms if they are missing. All stocks are sorted into deciles at the end of each month t based on either accruals or size calculated using the values from month t -1.The

23 _{Ayasdi is a machine learning company that allows investors to access alternative datasets like satellite data that}

allows investors to take informed bets on commodities by analyzing crop and weather data or estimate metrics like Sales at retailers by monitoring the number of cars in their parking lots.

33

portfolios are constructed using an value-weighted approach, where the weights are the market value of equity of every firm in a portfolio. When constructing monthly variables like accruals that rely on quarterly items from Compustat Point-in-Time, this paper uses the values reported for the last quarter for which reported financial statements are available.

This paper uses the auditor opinion of internal control quality under SOX section 404 as Ashbaugh et al. (2008) highlight that it is an unbiased signal about the effectiveness of a firm’s internal control system from an independent third party. Additionally, Rice and Weber (2012, 2013), highlight that firms have weak incentives to report internal control weaknesses when they exist and only a third of the firms may not have reported internal control weakness in their original disclosure for that period. This has implications for any paper, including this one that examines firms with weak internal controls as they will not have access to the entire population of firms with weak internal controls at any given point of time without being exposed to look- ahead bias. Accordingly, firms that file a restatement for a period are excluded from the sample for that period.

This paper follows Green et al. (2011) and constructs AGG_RANK, AGG_DIFF, AGG_RELPERSIST, AGG_IVOL, AGG_PRC, AGG_TRADING and TIME. All of these variables are computed every month t using data from month t-1. TIME =0.01 in January 2004 and increases by 0.01 through December 2017. AGG_TRADING is the market cap-weighted average of TRADINGit, which is the log of extreme accruals firm i’s average daily trading for the month t-1. Idiosyncratic volatility is chosen as one of the explanatory variables as Green et al. (2011) is use it in their model, which finds that the magnitude of the accruals anomaly has shrunk over time. AGG_IVOLit is the market cap-weighted average of IVOLit, which is the log of the standard deviation of residuals from a times series market model regression of the daily

34

returns of extreme accruals firm i’s stock on the CRSP value-weighted index over the month t-1. This paper calculates IVOLit using a Fama-French five-factor model regression instead of using the market model, which was used by Green et al. (2011). Additionally, Stambaugh, Yu, and Yuan (2015) find that stocks with higher idiosyncratic risk have stronger anomaly returns as idiosyncratic risk deters price-correcting arbitrage. Mashruwala et al. (2006) lend support to Stambaugh and Yuan’s (2015) findings as they provide evidence that the accruals anomaly is stronger among stocks with high idiosyncratic volatility. AGG_PRCt is the equally weighted

average of PRICEit, which is the log of extreme accruals firm i’s average price for the month t-1.

AGG_DIFF is the difference between the market cap–weighted average of accruals in the highest accruals decile and the market cap–weighted average of accruals in the lowest accruals decile sorted on accruals from month t-1 every month t. AGG_RELPERSIST is βACCRUALS /

βCASHFLOWwhere βACCRUALSand βCASHFLOW are the estimated coefficients in the following yearly

cross-sectional regression: 𝐼𝐼𝐼𝐼𝑖𝑖,𝑡𝑡+1 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡+1=𝑁𝑁𝑁𝑁𝑇𝑇𝑖𝑖+βACCRUALS𝑖𝑖 × 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖𝑡𝑡 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖𝑡𝑡 + βCASHFLOW𝑖𝑖× 𝐴𝐴𝐴𝐴𝐴𝐴𝐶𝐶𝐶𝐶𝐴𝐴𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖𝑡𝑡 + 𝜀𝜀𝑖𝑖𝑖𝑖 (3)

where IB represents annual income before extraordinary items, CF represents annual operating cash flows, and ACCRUALS is measured as IB minus CF. AVGAT is average annual total assets, and INT is an intercept. The variables are winsorized at the 1% level every year to adjust for outliers.