Yes, You Can! Surprising Model Setting in Empirical Finance Studies

Yes! You! Can!

1. Do not include firm fixed effects in a staggered difference-in-difference (DiD) model

We follow Lel and Miller (2015) and do NOT include firm fixed effects, as our dependent variable has little time series variation within each firm.

2. Using two instrumental variables for only one endogenous independent variable

For the two-stage least squares regression, we follow Deng and Gao (2013) and Yonker (2017a)and use the percentage of clear days in the firm’s headquarter state (PctClearHQ) and thepercentage of clear days in the CEO’s home state (PctClearHome) as the instruments (IV). As people in general prefer sunny weather, firms in regions with more desirable weather can more easily attract talented CEOs from across the country and are thus less likely to hire locally.

3. Call a regression table using Δvariables “change regression” (indeed, not very surprising 😅)

Another approach to establishing causality between CEOs’ locality and their firms’ myopic behavior is to conduct the tests within the same firm through a change analysis. To the extent that firms’ cultures and strategic orientations do not dramatically shift in a short period, a change in myopic behavior accompanying a change of CEO locality would suggest a causal relationship.

4. Address the Chen, Hribar, and Melessa (2018) incorrect inferences of 2-stage models by just adding first stage variables

For the abnormal accrual specification in Equation (2), we follow the approach of Chen et al. (2018) and include a vector c containing all the independent variables from the first-stage accruals model in Equation (1). This procedure is used to account for potential econometrical issues related to the use of residuals from a first-stage regression model as a dependent variable in a second-stage regression model. All error terms are clustered at the firm level.

5. Using industry averages as the instrumental variable

We select lagged industry averages of each customer concentration measure as our instrumental variables. Specifically, we use two-year and three-year lagged industry averages of the particular concentration measure. We calculate industry averages based on the supplier’s 3-digit SIC industry and year and exclude the supplier’s customer concentration from this calculation. For example, the instruments for whether a supplier has a major customer are the fraction of firms in the supplier’s 3-digit SIC industry that report relying on at least one major customer two years ago and three years ago.

Average industry customer concentration should meet the relevance condition because it is a good proxy for the structure of a supplier’s industry and will therefore be highly correlated with an individual supplier’s customer-base structure. Further, to the extent that industry customer concentration is correlated with industry risk and a supplier’s financing costs are related to industry risk, industry average concentration measures are unlikely related to the individual supplier’s financing costs after controlling for the individual supplier’s risk. By using lagged industry average values, it is also less likely that industry averages will be correlated with a supplier’s current cost of equity because investors price future risk not historical risk. Thus, industry average customer concentration measures likely satisfy the exclusion restriction. Yet, because the industry structure of customers is likely persistent through time, these lagged values should still be related to a supplier’s current customer concentration.

6. Use lagged Y in baseline model but not in the DiD model

7. Use a entropy balanced sample throughout the paper

An assumption in our research design is that exposed and unexposed observations are similar other than being exposed to bank failure through their auditor. To address potential concerns that our results are driven by different characteristics of exposed and unexposed bank-quarters, we utilize entropy balancing. We choose entropy balancing for two primary reasons. First, entropy balancing allows us to achieve a high level of covariate balance between the treatment and control groups by balancing across three moments (McMullin and Schonberger 2020). In contrast with other common matching methods, such as nearest neighbor matching, entropy balancing utilizes a reweighting scheme to achieve balance, allowing us to retain all observations in our sample and, thus, avoiding valuable loss of information. Second, the direct adjustment to weights based on sample moments eliminates the need for iterative searching over propensity score models, which can bias matching outcomes (Hainmueller 2012).

8. Publish a purely mathematical modeling paper on a Accounting journal

• Here are some examples:

Financial Reporting Discretion, Managerial Myopia, and Investment Efficiency

ABSTRACT. We explicitly model financial reporting discretion and earnings management in an investment setting where managers have incentives to behave myop

https://publications.aaahq.org/accounting-review/article/97/2/291/4369/Financial-Reporting-Discretion-Managerial-Myopia

Contemporary Accounting Research | CAAA Journal | Wiley Online Library

This study investigates the real effects of management communication, specifically of forecasts or earnings guidance, on investment. Managers can signal the strength of their projects through accurac...

https://onlinelibrary.wiley.com/doi/full/10.1111/1911-3846.12820