The sharpest tool in the shed: IPO financial statement management of STEM vs. non-STEM firms

STEM is an acronym for the fields of science, technology, engineering and math. The acronym arose in common use shortly after an interagency meeting on science education held at the U.S. National Science Foundation. Discussion of STEM-related programs has become a presidential priority because too few college students are pursuing degrees in these fields (WIKI Definition).

“Old-economy” stocks represent large, well-established companies that participate in more traditional industry sectors and have little investment or involvement in the technology industry, whereas “new-economy” stocks are heavily involved in the technology sector, and the more successful companies can build value at markedly higher growth rates. http://​www.​investopedia.​com/​ask/​answers/​05/​061305.​asp#ixzz4BUXWVVnp

Zimmerman (2015, p. 488) elaborates: “To be clear, accounting net income is a poor measure of the manager’s performance in firms with substantial growth opportunities … In these firms, managers can be adding significant value to the firm yet accounting earnings report losses. So, in growing twenty-first century human-capital-intensive firms, accounting net income is a poor motivator of the key stakeholders. Rather than utilizing accounting net income, other accounting numbers derived from the audited financial reports can prove constructive in designing incentive contracts in twenty-first century firms, such as revenues, cash flows, cost reduction targets, working capital, and so forth.”

The following quote from Dechow et al.’s (2010, p.390) review article helps motivate our study: “When making choices that affect reported earnings, a manager’s objective function can include multiple, and perhaps competing, objectives. … [an] interesting path for future research is to examine choices about the portfolio of accounting choices, specifically within the context of meeting multiple objectives.” Our paper aims to advance this suggested path, namely to examine a portfolio of accounting choices made by the manager and to present evidence on the tools managers of IPO firms use depending on the nature of the firm.

The timeline of Year 0 and Year −1 is depicted in Figure 1 and discussed in Section 4.

Marquardt and Wiedman (2004) show that firms exercise discretion over different components of earnings when they have different objectives related to earnings (i.e., equity offerings, management buyouts, and avoiding earnings decreases).

The SIC codes for this sector are 355, 357, 360–69, 381, 382, 384, 386, 481, 737, and 873 (except for 8731).

In untabulated analysis, we partition the sample firms along the lines of profit versus loss and rerun our tests. Although there is much to be gleaned from this analysis, we do not tabulate these results to keep the paper focused. Yet this alternative partitioning generates similar results, as we find a use of discretion when the accounting item is priced.

For robustness, we also include book value (BV) in model (1), consistent with the framework laid out by Ohlson (1995), among others. However, the large flow of capital from the proceeds of the IPO creates a very strong mechanical association (Aggarwal et al. 2009). Nonetheless, it does not change our results. Alternatively, we added lagged BV, and, although it did not change the results on our primary variables of interest, in some specifications the coefficient on the BV itself was negative, which is counterintuitive. Upon further analysis, we determined that the large number of IPO firms with negative BV was distorting the results (47% of our sample have negative BV). Thus we decided that, although the Ohlson (1995) framework is useful for steady-state firms, the IPO setting was not conducive to the inclusion of BV in the valuation model.

As an alternative procedure, we follow Kothari et al. (2005) and match each IPO firm with a non-IPO firm on ROA, industry, and year and then subtract the matched firm’s residual from that of the IPO firm to derive the performance-adjusted discretionary item for the IPO firms. Results using this procedure are qualitatively similar to those using the Armstrong et al. (2016) procedure. A similar procedure where we additionally match on book-to-market ratio yields similar results, except that the sample size decreases.

Discretionary sales may include channel stuffing, bill and hold sales, and revenues recognized using aggressive or fraudulent application of GAAP. This measure captures both accrual-based and real activity sales management.

Stubben (2010) also includes the interactions of change in sales with log age and with log age squared as explanatory variables and uses the number of years since the firm first appeared on Compustat annual file to proxy for firm age. Unfortunately, this measure cannot be used in our study because it will eliminate the variability in the firm age of IPO firms or simply flag IPO firms. When we use in the model firm age using the time from founding year, both variables are insignificant and the size of our sample drops. Therefore we exclude those variables in equation (2).

As a robustness test to address the potential endogeneity of R&D and Tobin’s Q, we replace Tobin’s Q with sales growth in the following two years. The estimated discretionary R&D levels are qualitatively similar. However, this specification introduces a look-ahead bias in the return tests and causes a reduction in the sample size. In the analysis reported in Section 6.5, where we examine the use of discretion in the pre-IPO year (Year −1), we use this specification because market value of equity (and consequently Tobin’s Q) is unavailable before the company becomes public.

To address the concern that R&D might be a life-cycle variable, as a robustness check, we match the IPO firms with non-IPO firms on each of the following: 1) lagged R&D, 2) Tobin’s Q, and 3) lagged R&D and ROA in order. We obtain very similar results using these alternative matching procedures.

Some studies use total accruals, as opposed to current accruals, to calculate discretionary accruals (e.g., Fan 2007; Armstrong et al. 2016). For robustness, we replicate all our tests using total accruals and obtain similar results.

A caveat to this analysis is that all of the models described above that we use in the paper properly specify discretion from nondiscretion. Going back to Bernard and Skinner (1996), this issue has plagued these models and despite decades of research, cannot be completely addressed. Investors may simply overinvest, relative to what our models would predict (maybe because they are going through a rapid growth stage), but there is no foul play happening. Another possibility is that managers are simply investing more than our models would predict, which we interpret as over-investment. Although we cannot rule out this possibility, some of our testing leads us to conclude that is not the predominant case. If it were, then we believe this would add noise to our discretionary components and possibly work against us finding results. However, as with all studies of this nature, it must be acknowledged that we may have mis-categorized legitimate investment as discretionary.

If a company delists during the 12 months after the reporting, we include the delisting return in the abnormal return calculation and re-invest the proceeds in a zero abnormal return portfolio. If the monthly delisting return is missing, then we use the delisting return from the daily return file. If the delisting return is also missing from the daily return file, we use the average delisting return for the relevant category.

Data was hand-collected because of a high error rate in the SDC data for that period. We are thankful to Professor Ljungqvist for sharing his data with us.

The number of observations in the various tests is smaller and varies according to data availability.

To test for a statistical difference between the coefficients of STEM and non-STEM firms, we estimate a regression using the entire sample with interaction variables for the STEM firms (e.g., LOGSGR*STEM_dummy, etc.). We find that positive earnings are significantly higher for the non-STEM firms, while sales growth is significantly higher for the STEM firms, whereas negative earnings are not significantly different between the two groups. The coefficient on R&D while stronger for STEM firms, is not significantly different than that of the non-STEM firms. However, when we also consider proportion of the price that is explained by R&D and the marginal effect on the price (Figure 2 and Table 2 Panel B analyses), the difference between the two groups is strongly pronounced for all three variables.

As a robustness test, we re-estimate model (1) after partitioning PEBXI_noRD, NEBXI_noRD, LOGSGR, and RD into their discretionary and nondiscretionary components. We find that the market appears to price both components, lending support to the argument in the literature that IPO managers have incentives to manage earnings since the market seems to price the discretionary components as well.

For robustness, we plotted the levels of our discretionary items from Year −1 to +5 (untabulated). We find that the levels of abnormal items converge quickly toward zero over the following year or two.

The finding of overinvestment in R&D during the IPO differs from the conclusion in Darrough and Rangan (2005) of underinvestment in R&D by IPO firms. For a much smaller sample of 243 IPOs during 1986–1990, the authors find a negative association between the change in R&D spending and managerial equity sales. They infer that managers believe that, for IPO firms, investors place more emphasis on current earnings and less emphasis on R&D and therefore cut on R&D. Note that they do not test directly for underinvestment but infer it from the association with insider sales. In contrast, in our study, we directly test the use of discretion over R&D and find that managers of STEM firms inflate R&D because of valuation incentives. It is also worth noting that there is a very little overlap between the sample period in Darrough and Rangan (2005) and our study.

As a robustness test, we add cash flows to the stock returns models. However, accruals and cash flow from operations are highly negatively correlated (Dechow 1994), and we also find discretionary earnings to be negatively correlated with cash flow. This may create high collinearity in the regressions. We also find that discretionary sales and discretionary R&D are negatively correlated with cash flows. To avoid multicollinearity, we first regress cash flows on the three discretionary items. We then include the residual of cash flows in the return regressions. The results show that the cash flow item is significant, and that our main findings remain the same.

We also find a negative association between discretionary earnings, discretionary sales, and discretionary R&D and the long-term operating performance of the IPO firms. For sake of brevity, these results are untabulated.

We also examine the association between the discretionary items and insider trading. We follow Armstrong et al. (2016) and measure insider trading as the difference between insider purchases and insider sales scaled by the sum of the two (John and Lang 1991), focusing on open market transactions by senior officers and officer directors. To alleviate the effect of small trades, we only consider firms with insider trading of at least 1000 shares during that period. We calculate the insider trading over the 12 months from the month of the release of the financial statements for Year 0. We find some evidence of an association between the discretionary measures and insider trading (untabulated). Discretionary sales are weakly negatively associated with insider trading for the entire sample. The association is highly significant for the STEM firms that have the greater discretionary sales, and insignificant for the non-STEM firms. We do not find a significant association between insider trading and neither discretionary earnings, nor discretionary R&D. The association of discretionary sales with insider trading is consistent with recent evidence by Dechow et al. (2016) that insiders profit by the use of accounting revenue recognition letters.

Hall et al. (2005) find that each additional citation per patent increases the market value by 3%. They also show that, while the average and below average patent quality does not affect market value, patents with more than 20 citations boost market value by 50–75%. Trajtenberg (1990) also shows that patent citation data is highly skewed with almost half the patents never cited.

We set discretionary R&D to 0 for firms with no R&D expenditures.