Product Life Cycle Lengths and Real Activities Management

Abstract

We examine how short product life cycle lengths constrain real activities management among a group of manufacturing firms. Companies with short-cycle products face fast product obsolescence and constant pressure to develop and market new products. Fast product obsolescence increases the cost of overproduction, while constant pressure for new product development increases the cost of opportunistic spending cuts. We predict and find evidence that in general, short-cycle firms do not beat earnings thresholds through overproduction and discretionary spending cuts as extensively as long-cycle firms do. However, since the Sarbanes-Oxley Act came into effect, short-cycle firms have increasingly overproduced, and there is no longer a difference in overproduction between short- and long-cycle firms. Moreover, the presence of major customers attenuates the relationship between product life cycle lengths and real activities management, while a large market share and high fixed-assets intensity both accentuate the relationship. Finally, short-cycle firms are also less likely to use LIFO (last-in, first out) liquidation to inflate earnings than long-cycle firms.

Keywords

real activities management overproduction product life cycle lengths absorption cost accounting discretionary spending cut LIFO liquidation

Introduction

Real activities management (RAM) is harder to detect by external stakeholders than accruals manipulation (Chen et al., 2020; Cohen et al., 2008). However, it is often harmful to long-term firm value (Cohen & Zarowin, 2010). To better detect and prevent RAM, it is important to recognize various constraints on RAM. Such constraints can result from a firm’s operating characteristics and corporate governance environment. Researchers have identified several external and internal corporate governance factors that constrain RAM (e.g., Bushee, 1998; Chen et al., 2015; Cheng et al., 2016; Zang, 2012). Recently, Huang et al. (2020) have suggested the threat of litigation as another constraint. In this article, we examine whether short product life cycle lengths act as an operating constraint on RAM.

Product life cycle length is a well-developed concept in the operations management literature. It describes the duration from the time a product is introduced to the market to the time it becomes obsolete or is replaced by a new version. How long new products remain viable in the marketplace varies widely from industry to industry. Some industries have very short product life cycle lengths due to fast-changing consumer preferences and rapid technology advancement (e.g., watches, consumer electronics, and personal computers).¹ Companies in these industries face unique challenges such as fast product obsolescence and difficulty in demand forecasting (Kurawarwala & Matsuo, 1996). RAM tools such as overproduction and discretionary spending cuts are likely more costly for such companies.

Prior studies have established that managers overproduce to achieve important earnings objectives (e.g., Roychowdhury, 2006). We argue that overproduction is not a desirable earnings management choice for short-cycle firms because of higher inventory risk. Short-cycle firms are more likely to end up with obsolete inventory if they overproduce, and either carrying obsolete inventory or writing it off can seriously harm companies’ financial performance (Stout & Hanson, 1989). Long-cycle firms, on the contrary, have more flexibility to sell overproduced items because their products are viable in the market for a longer time. In other words, it should be more costly for short-cycle firms to overproduce. Our first question examines whether short-cycle firms are less likely to achieve important earning targets through overproduction than firms with longer product life cycle lengths.

The literature also suggests that managers often cut discretionary spending—such as research and development (R&D) expenses and selling, general, and administrative (SG&A) costs—in their efforts to meet earning targets (e.g., Bushee, 1998; Roychowdhury, 2006). We argue that short product life cycle lengths also make such spending cut more costly. Industries with short-cycle products are driven by consumers’ ever-changing tastes and rapid technological advancement. Short-cycle firms are under constant pressure to develop new products that respond to shifts in customers’ preferences or technical innovations. Deviating from an optimal R&D strategy can be more costly for them. Regarding marketing expenses, short-cycle firms must incur additional costs beyond regular promotional spending to defend existing products whenever competitors introduce new products to the market, which happens more often than for firms with longer product life cycle lengths. In other words, cutting SG&A expenditure is more costly for short-cycle firms. Our second question, therefore, examines whether short-cycle firms also cut discretionary spending to a lesser extent than their long-cycle counterparts to meet earnings thresholds.²

As an extension to our main research questions, we examine whether the short product cycle lengths become less effective in constraining RAM in the post-SOX period. The literature suggests that managers tend to choose earnings management alternatives with lower costs (e.g., Graham et al., 2005; Zang, 2012). As the regulatory environment becomes more stringent post-SOX, accrual manipulation is subject to more scrutiny (e.g., Cohen et al., 2008). Even if operating constraints impose sizable costs on RAM pre-SOX, RAM is likely to become more desirable when the costs of alternatives are increased. Therefore, we examine whether short product life cycle lengths become a less effective constraint on RAM after SOX.

Using data on U.S. manufacturing firms from 1976 to 2017, we examine the above questions. We measure firms’ product life cycle lengths based on product obsolescence induced by technological changes. If the underlying technology changes, products adopting the outdated technology will be obsolete. Following Bilir (2014), we measure this dimension of product life cycle lengths using leading citations of patents in an industry. We arrive at several findings. First, in the full sample, short-cycle firms do not overproduce or cut discretionary spending as extensively as long-cycle firms do. This is consistent with the notion that short product life cycle lengths serve as a natural constraint on RAM. Second, short product life cycle lengths constrain overproduction before SOX. After SOX, short-cycle firms have increasingly engaged in overproduction, to the extent that differences between short- and long-cycle firms are no longer discernable. We also identify cross-sectional factors that either strengthen or weaken the predicted relationships. For example, the presence of major customers weakens the relationship between product life cycle lengths and RAM. On the contrary, large market share and high fixed-assets intensity accentuate the relationship.

We conduct several additional robustness checks. First, we compare the subsequent-year discretionary spending for suspect short-cycle versus suspect long-cycle firms. We find evidence that short-cycle firms tend to reverse spending cuts more than long-cycle firms. This evidence is again consistent with the notion that opportunistic spending cuts are more costly for short-cycle firms, providing further support for our main hypotheses. In the main test, we measure the product life cycle lengths with PLC, a rank variable of product life cycle lengths in years (T). As a robustness check, we use T to measure this construct, and the results are qualitatively the same. In unreported tests, we use alternative measures for key control variables such as firm life cycle stage, and we also control industry capacity utilization to make sure our overproduction results are not driven by firms’ capacity constraints. Finally, we use the last consensus analyst forecast calculated before the end of the fiscal year as well as the last consensus forecast before the annual earnings announcement. Inferences remain the same.

We contribute to the literature in several ways. First, our article suggests that short product life cycle lengths act as an operating constraint on RAM. Prior literature has identified several corporate governance factors and the threat of litigation risk as constraints (e.g., Bushee, 1998; Chen et al., 2015; Cheng et al., 2016; Huang et al., 2020). Our article suggests that a firm’s operating characteristics can also limit managers’ abilities to engage in RAM. Our finding should be relevant to both researchers and investors who are interested in detecting and preventing RAM.

This paper should also be of interest to regulators. There has been a growing literature on the effects of SOX on earnings management (e.g., Hossain et al., 2011). Our results add to this line of research. Specifically, short product cycle lengths were effective in constraining RAM pre-SOX, but this constraint has been weakened as RAM became more popular in the post-SOX era. This finding highlights the unintended consequences of SOX (e.g., Cohen et al., 2008; Eli & Cohen, 2009). It echoes the view of Dechow et al. (2010) that “the overall effect of SOX on the decision usefulness of earnings is ambiguous.”

Third, our study contributes to the understanding of how operating decisions are made in short-cycle firms. While there is no shortage of studies on the optimal production and inventory levels in the operations management literature, most papers on this topic rely on anecdotal evidence or theoretical models (e.g., Kurawarwala & Matsuo, 1996; Martinez-de-Albeniz & Simchi-Levi, 2009). We add to this literature by providing large-sample evidence on how product life cycle lengths affect firms’ production and discretionary spending decisions when firms have income-increasing incentives.

Our article is structured as follows. In the next section, we discuss hypotheses development. In section “Empirical Design and Sample Selection,” we describe the methodology and discuss sample selection. In section “Empirical Results,” we present empirical results and robustness test results. We summarize and conclude in the final section.

Hypotheses Development

In a survey by Graham et al. (2005), most financial executives acknowledge that they are willing to manipulate firms’ real activities to meet important earnings targets, even if such behavior often reduces long-term firm value. Managers prefer RAM because it is harder to detect than accrual manipulation (Chen et al., 2020; Cohen et al., 2008). Not surprisingly, researchers and investors are interested in ways to detect and constrain RAM. Bushee (1998) and Zang (2012) provide evidence that the presence of sophisticated institutional investors deters RAM. Chen et al. (2015) suggest that CEO contractual protection helps to reduce RAM such as R&D spending cuts. Cheng et al. (2016) demonstrate that the horizon and influence of key subordinate executives deter RAM, consistent with effective internal corporate governance reducing real activities manipulation by CEOs. Finally, Huang et al. (2020) suggest that the threat of litigation also discourages managers from engaging in RAM.

In this paper, we examine whether short product life cycle lengths make RAM particularly costly and hence serve as an operating constraint on RAM. Product life cycle lengths vary greatly by manufacturer. Some industries enjoy long product life cycle lengths. For example, the life cycle length of a vacuum cleaner is more than a decade, and “you get breathing space [to optimize and improve the product].”³ However, some other industries have much shorter product cycle lengths (e.g., watches, consumer electronics, and personal computers). These industries follow consumers’ rapidly changing preferences, and products often sell for 1 or 2 years before new products are introduced. Mr. Hasebe, president of Sony Logistics, has commented that “inventory is a business risk in a fast-changing market.”⁴

Under the absorption costing required by Generally Accepted Accounting Principles (GAAP), fixed overhead is spread over units produced. Producing beyond sales demand reduces the unit production cost and generates inflated profits. Existing studies find evidence that firms overproduce to beat earnings targets (e.g., Roychowdhury, 2006). We argue that short-cycle firms are more concerned with inventory risk than their long-cycle counterparts. This is because overstocked products are more likely to become obsolete in short-cycle industries. To sell obsolete items, companies need to offer steep discounts, which puts pressure on their cash flows (Stout & Hanson, 1989). Moreover, offering discounts is not always straightforward; it runs the risk of diluting brand images.^5,6 When steep discounts still fail to help sell outdated products, companies must write off the inventory which negatively affects their financial performance.⁷

Taken together, short-cycle firms are more concerned with inventory risk. Compared to long-cycle firms, they face much higher economic costs of manipulating earnings through overproduction. The literature suggests that managers usually have several earnings management alternatives to choose from, and they tend to choose those with lower costs (Cohen et al., 2008; Graham et al., 2005; Zang, 2012). If short-cycle firms need to bridge earnings gaps, we argue that they will not engage in overproduction as much as their long-cycle counterparts. Our first hypothesis is as follows:

Hypothesis 1 (H1): Short-cycle firms do not beat earnings thresholds through overproduction as much as long-cycle firms do.

Next, we examine short-cycle firms’ tendency to cut discretionary spending to achieve financial reporting objectives. Following Roychowdhury (2006), we measure discretionary spending by R&D expenditure, advertising, and SG&A expenses. Because of the short product life cycle lengths, short-cycle firms are under constant pressure to develop new products.⁸ The ability to bring product ideas to fruition ahead of competitors determines their economic success. Moreover, most of the new products in short-cycle firms are technology dependent. Delays in product development increase the chance that product ideas become obsolete as new technology emerges. Given the constant pressure to innovate, short-cycle firms likely find it more costly to cut spending on research and new product development than long-cycle firms.

As to the spending on product promotion, generous marketing and distribution support is often vital for the commercial success of new products (Abedi et al., 2014). This is because financial resources are necessary to increase consumer awareness and to build the reputation of the new products. Moreover, companies must incur defensive marketing costs. Hauser and Shugan (2008) point out that whenever new products are introduced, incumbents must defend their positions in the market. Since new product introduction occurs more frequently in industries that sell short-cycle products than in those that sell long-cycle products, short-cycle firms need to incur more expenditure to defend their existing products from new ones. In other words, we argue that short-cycle firms find it more costly to cut advertising and SG&A spending to remain competitive in the market. Therefore, our second hypothesis is as follows:

Hypothesis 2 (H2): Short-cycle firms do not beat earning thresholds by cutting discretionary spending as much as long-cycle firms do.

Next, we discuss several factors that may either strengthen or weaken the relationships documented in H1 and H2. The first factor we consider is the presence of major customers. Existing studies suggest that the implicit contracts between companies and their supply chain partners induce better-quality financial reporting (e.g., Bauer et al., 2018; Hui et al., 2012). Since major customers care about companies’ information risk, companies with major customers face higher costs relating to reputation damage from earnings manipulation than those without major customers. If major customers make both long- and short-cycle firms refrain from RAM, the relationships we observe in H1 and H2 will be weakened in the subsample with major customers. Therefore, our third hypotheses are as follows:

Hypothesis 3a (H3a): The presence of major customers attenuates the relationship between product life cycle lengths and overproduction.

Hypothesis 3b (H3b): The presence of major customers attenuates the relationship between product life cycle lengths and discretionary spending cuts.

We argue that firm life cycle stage is also a dimension that affects both firms’ overproduction and discretionary spending decisions. Dickinson (2011) classify firms into five life stages based on their cash flow patterns, that is, introduction, growth, mature, shake-out, and decline. Firms in the early stages of their life cycles have had less time to develop distribution networks than their later-stage counterparts. This factor likely makes early-stage, short-cycle firms even more reluctant to overproduce; because it is more difficult for them to sell obsolete items via their less-developed distribution channels. Similarly, early-stage firms are at the stage of building product and company awareness. They may have more to lose if they deviate from optimal business practices. If being in the early stages increases the cost of discretionary spending cuts especially for short-cycle firms, the relationship we observe in H1 and H2 will be more salient in the subsample of early-stage firms. Our next hypotheses are as follows:

Hypothesis 4a (H4a): Being an early-stage firm accentuates the relationship between product life cycle lengths and overproduction.

Hypothesis 4b (H4b): Being an early-stage firm accentuates the relationship between product life cycle lengths and discretionary spending cuts.

Zang (2012) suggests that firms in market-leader positions enjoy many competitive advantages over their peers, such as economies of scale, and bargaining power with suppliers and customers, and so on. Because of the various advantages, managers in such firms perceive RAM as less costly. It follows that the relationship between product life cycle lengths and RAM is likely stronger among market leaders. Market followers worry about further eroding their competitive positions by deviating from the optimal business strategy. Such concerns discourage both long- and short-cycle market followers from engaging in extensive RAM. By contrast, market-leader status takes away the extra competitive pressure. Thus, product life cycle lengths are more likely to have a salient effect on the perceived costliness of RAM among market-leading firms. Therefore, our next hypotheses are as follows:

Hypothesis 5a (H5a): Market-leader status accentuates the relationship between product life cycle lengths and overproduction.

Hypothesis 5b (H5b): Market-leader status accentuates the relationship between product life cycle lengths and discretionary spending cuts.

Finally, we discuss a factor that specifically impacts overproduction. Higher fixed-asset intensity allows a firm to be more effective in inflating earnings through overproduction (Gupta et al., 2010). In a subsample with low fixed-asset intensity, although long-cycle firms face lower costs and have more incentives to overproduce than short-cycle firms do, their abilities to inflate earnings are limited. By contrast, high fixed-asset intensity makes overproduction an effective earnings management method to reduce the cost of goods sold and inflate earnings. Among such firms, product life cycle lengths (and concern about inventory risk) are more likely to have a meaningful impact on firms’ overproduction decisions. Hence, we make the following prediction:

Hypothesis 6 (H6): Having high fixed-asset intensity accentuates the relationship between product life cycle lengths and overproduction.

Empirical Design and Sample Selection

Sample

We study U.S. manufacturing firms (i.e., Standard Industrial Classification [SIC] codes 20–38) over the period 1976 to 2017. We focus on manufacturing firms for two reasons. First, manipulation with overproduction is most relevant for manufacturing firms (Roychowdhury, 2006). Second, our key metric, PLC, is based on corporate patent and citation information, and manufacturing industries have active patenting activities. We choose the sample period from 1976 to 2017 because the National Bureau of Economic Research (NBER) U.S. Patent Citation Data File that is used to construct PLC starts coverage from 1976. We make use of the Compustat database for inventory, R&D expenses, SG&A expenses, and other accounting data. We focus on two earning targets: consensus analyst forecasts and last years’ earnings, and obtain analyst forecast information from the IBES database. Finally, for cross-sectional tests on major customers, we obtain supplier–customer information from the Compustat Segment database.

Measure of Product Life Cycle

Product life cycle length describes the period from when a product is introduced to the market to the time it becomes obsolete or is replaced by a new version. Product obsolescence occurs when technology embedded in a new product replaces the existing technology underlying a current product. For example, smartphones made traditional personal digital assistants and cellular phones largely obsolete. Sometimes, products become difficult to sell with the introduction of a new version or upgrade, although the new version has the same underlying technology but boasts a new design. For example, Apple Inc. introduces new iPhone models every year. This does not make the old model obsolete immediately, but the new model tends to gradually replace the old version.⁹ Since we do not have data on the sales duration of different product versions and models, we measure product life by the economic life of embedded technology.

Following Bilir (2014), we measure the economic life of a technology as the length of time for which the underlying patent continues to be cited by subsequent patents. A longer time lapse between a patent’s grant date and the most recent citation it received (i.e., forward citation lag) indicates a longer economic life of the cited patent and longer life spans of products that adopt this technology. In specific terms, one calculates the average forward citation lag for each technology class reported in the NBER U.S. Patent Citation Data File (1976–2006) and then matches it to three-digit SIC codes using U.S. Patent and Trademark Office (USPTO) concordance data. In other words, the PLC metric is industry-specific.¹⁰ Within our final sample with the necessary financial data, we convert the product life cycle lengths in years (T) to a rank-order variable (1–55) as our final measure of product life cycle lengths (PLC).¹¹ We then make the assumption that PLC remains stable throughout the entire sample period. This assumption makes it possible to test our hypotheses over the entire 40-year period.

Research Models

We capture overproduction by calculating unexpected production costs (UnexpProd). Following existing literature (Roychowdhury, 2006; Srivastava, 2019), we measure UnexpProd as the residual value estimated from the following regression:¹²

\begin{matrix} \frac{Pro d_{t}}{T A_{t - 1}} = α_{0} + α_{1} \frac{1}{T A_{t - 1}} + α_{2} \frac{Sale s_{t}}{T A_{t - 1}} + α_{3} \frac{Δ Sale s_{t}}{T A_{t - 1}} + α_{4} \frac{Δ Sale s_{t - 1}}{T A_{t - 1}} \\ + α_{5} \frac{Sale s_{t + 1}}{T A_{t - 1}} + α_{6} {MTB}_{t} + ϵ_{t}, \end{matrix}

where Prod is the sum of the cost of goods sold and change in total inventory, Sales is sales revenue, TA is total assets, and MTB is calculated as the market value of equity divided by book value of equity. We regress the above model for each industry year, where the industry is based on two-digit SICs. We require a minimum of 15 observations in each regression.

We measure the reduction in discretionary spending by calculating the unexpected portion of discretionary spending (UnexpDisExp). Following Roychowdhury (2006) and Srivastava (2019), we run the following regression for each industry year:

\frac{DisEx p_{t}}{T A_{t - 1}} = α_{0} + α_{1} \frac{1}{T A_{t - 1}} + α_{2} \frac{Sale s_{t - 1}}{T A_{t - 1}} + α_{3} \frac{Sale s_{t + 1}}{T A_{t - 1}} + α_{4} {MTB}_{t} + ϵ_{t},

where DisExp is the sum of R&D expenses, advertising expenses, and SG&A expenses, and industry is again defined by two-digit SIC codes. We measure UnexpDisExp as the residual value estimated from the above regression.

Our first hypothesis examines the prevalence of overproduction among short- versus long-cycle firms with incentives to inflate earnings. We use the following regression model to test this hypothesis:

\begin{matrix} UnexpProd = β_{0} + β_{1} PLC + β_{2} EM + β_{3} PLC * EM + β_{4} ZScore + β_{5} ETR + \\ β_{6} Big 5 + β_{7} NOA + β_{8} Cycle + β_{9} LnMVE + β_{10} ROA + β_{11} LnFirmAge + \\ β_{12} ForSeg + β_{13} RDInt + Year Dummies + ε . \end{matrix}

(1)

In regression model (1), the variable of interest is an interactive term of PLC and EM (PLC*EM). Following prior literature (e.g., Graham et al., 2005; Payne & Robb, 2000), we focus on two earnings thresholds, namely consensus analyst forecasts and last year’s earnings. Accordingly, we construct two earnings management (EM) variables—Beat and Increase. Following Zang (2012), we code Beat as 1 if the actual EPS is equal to or greater than consensus forecast by no more than 1 cent, and 0 if the firm-year misses or beats analyst consensus forecasts by more than 5 cents.¹³ Increase is coded as 1 if the change in net income from last year to this year falls in the bin immediately to the right of zero, and 0 otherwise. Since a higher PLC value indicates a longer product life cycle, we predict positive coefficients for both PLC*Beat and PLC*Increase.

In Hypothesis 2, we argue that short-cycle firms are less likely to cut discretionary spending for financial objectives. We run the following regression for this test:

\begin{matrix} UnexpDisExp = β_{0} + β_{1} PLC + β_{2} EM + β_{3} PLC * EM + β_{4} ZScore + \\ β_{5} ETR + β_{6} Big 5 + β_{7} NOA + β_{8} Cycle + β_{9} LnMVE + β_{10} ROA + \\ β_{11} LnFirmAge + β_{12} ForSeg + β_{13} RDInt + β_{14} HHI + Year Dummies + ε . \end{matrix}

(2)

In model (2), the variable of interest is again PLC*EM. As more negative values of UnexpDisEXP correspond to greater spending cuts, we predict negative coefficients for PLC*EM. We conduct cross-sectional tests of H3 to H6 using subsample tests and will discuss the methods together with test results in the section of Empirical Results.

Descriptive Statistics

Panel A of Table 1 presents the names of the top and bottom five three-digit industries by PLC rank, as well as the number of observations in each industry. The five industries with the shortest product life cycle lengths have an average T of 7.32 to 8.76 years. On the contrary, firms with the longest product life cycle lengths have an average life cycle ranging from 10.40 to 13.5 years. Next, we present the sample industry distribution in Panel B of Table 1. Four industries dominate our sample: SIC 28 (Chemicals and Allied Products), 35 (Industrial and Commercial Machinery), 36 (Electronic and Other Electrical Equipment), and 38 (Measuring, Analyzing, and Controlling Instruments; Photographic, Medical and Optical Goods; Watches and Clocks). Together, firms in these four industries represent approximately 80% of the sample.

Table 1.

Sample Distribution by Industry.

Panel A: Product Life Cycle Lengths and Rank by Industry.
Industry description	PLC rank	SIC code	T (years)	No. of Obs.
Industries with the shortest product life cycles
Watches, clocks, clockwork operated devices, and parts	1	387	7.32	22
Computer and office equipment	2	357	8.32	4,928
Household audio and video equipment	3	365	8.51	467
Agricultural chemicals	4	287	8.51	238
Electronic components and accessories	5	367	8.76	5,268
Industries with the longest product life cycles
Cutlery, hand tools, and general hardware	51	342	10.40	359
Metal cans and shipping containers	52	341	10.61	113
Heating equipment, except electric	53	343	10.86	93
Electric transmission and distribution equipment	54	361	11.14	313
Coating, engraving, and allied services	55	347	13.50	75

Panel B: Sample industry distribution.
Industry description	SIC Code	No. of Ob	Percentage (%)
Food and kindred products	20	188	0.35
Textile mill products	22	138	0.26
Chemicals and allied products	28	11,982	22.60
Primary metal industries	33	1386	2.61
Fabricated metal products	34	1926	3.63
Industrial and commercial machinery	35	10,756	20.28
Electronic and other electrical equipment	36	12,656	23.87
Transportation equipment	37	3064	5.78
Instruments; photographic, etc.	38	10,930	20.61
Total		53,026

Table 1: Panel A reports the product life cycle lengths in years (T) and number of observations for the top and bottom five three-digit industries ranked by product life cycle lengths. Panel B reports the sample industry distribution, where industries are defined by two-digit SIC codes.

Table 2 presents descriptive statistics. The rank-order variable PLC, with a value range of 1 to 55, has a median of 14. For the two earnings management variables, Beat has a mean of 0.18, suggesting that 18% of sample firms report EPSs that just meet or beat analyst consensus forecasts by no more than 1 cent. Increase has a mean of 0.025. Approximately 2.5% of sample firms display small earnings increases from the previous year. Both mean and median Altman Z-scores for sample firms are above 3, suggesting solid financial positioning. However, the first quartile value is 1.98. A quarter of sample manufacturing firms appear to be headed toward bankruptcy. Big5 has a mean of 0.78. Three-quarters of sample firms are audited by Big 5 auditors. Sample manufacturers have an average (median) market value of $1.99 billion ($128 million). They are rather large compared to the average Compustat firms. On average, the sample firms have existed for 18 years. Finally, MjrCx (major customer) has a mean of 0.29, suggesting that a quarter of sample firms have at least one major customer.

Table 2.

Descriptive Statistics.

Variable	No. of Obs.	M	25%	Median	75%	SD
T (years)	53,026	9.24	8.97	9.12	9.59	0.54
PLC (in rank)	53,026	20.8	10	14	33	15.0
UnexpProd	52,955	−0.021	−0.18	−0.035	0.097	0.40
UnexpDisExp	48,371	−0.0099	−0.21	−0.042	0.14	1.03
Beat	20,115	0.18	0	0	0	0.39
Increase	53,026	0.025	0	0	0	0.16
ZScore	53,026	4.62	1.98	3.32	5.95	14.9
ETR	53,026	0.19	0	0.24	0.37	0.32
Big5	53,026	0.78	1	1	1	0.41
NOA	53,026	1.41	0.41	0.59	0.92	4.34
Cycle	53,026	123.0	69.3	118.3	178.9	168.9
MVE ($ in millions)	53,026	1991.5	27.2	127.7	730.6	6720.3
LnMVE	53,026	4.98	3.30	4.85	6.59	2.36
ROA	53,026	−0.13	−0.12	0.030	0.086	0.68
Firm age	53,026	17.8	8	14.0	24	13.0
LnFirmAge	53,026	2.62	2.08	2.64	3.18	0.74
ForSeg	53,026	0.80	1	1	1	0.40
RDInt	53,026	0.78	0.020	0.060	0.15	3.58
MjrCx	53,026	0.29	0	0	1	0.45
HHI	53,026	0.043	0.029	0.040	0.050	0.019
MKTShare	53,026	0.0037	0.000036	0.00022	0.0015	0.013
FAI	52,901	0.45	0.23	0.39	0.59	0.32
SOX	53,026	0.37	0	0	1	0.48

Table 2 presents the descriptive statistics. Variable definitions can be found in Appendix.

Empirical Results

Multivariate Test Results

Table 3 presents results on overproduction tests (including the main test H1 and the cross-sectional tests H3–H6). In Panels A and B, the earnings targets are consensus analyst forecasts and last year’s earnings, respectively. Panel A Column (A) presents the full sample (H1) results. The variable of interest, PLC*BEAT, has a coefficient that is positive and significant at 5%. There are strong evidence that short-cycle firms do not engage in overproduction to beat consensus forecasts as extensively as long-cycle firms do. Most control variables have predicted signs. For instance, Big5 is significantly positive. Due to the higher detection risk on accruals manipulation, clients of Big 5 auditors have greater overproduction. Cycle is negative and significant—firms with longer operating cycles have more flexibility with accrual manipulation and do not need to resort to overproduction as frequently. In addition, LnMVE is negative and significant. With better information environments, large firms engage in overproduction to a lesser extent than smaller firms. LnFirmAge is significantly positive, suggesting that mature firms are more likely to carry large inventory stock, because they have better distribution networks to sell overstocked products. Overall, the evidence from Column (A) provides support to H1.

Table 3.

The Impact of Product Life Cycle Lengths on Firms’ Overproduction.

Panel A: Beating analyst forecasts
	(A)	(B)	(C)	(D)	(E)	(F)	(G)	(H)	(I)
Variable	Full sample	W/O Mjr Cx	W/Mjr Cx	Early stage	Later stage	Low MKT share	High MKT share	Low FAI	High FAI
PLC	0.00077***(4.90)	0.00087***(3.68)	0.00058*(1.68)	0.0002(0.50)	0.0010***(3.52)	−0.000033(−0.078)	0.000026(0.13)	0.00075**(2.33)	0.00084***(3.89)
Beat	−0.070***(−6.04)	−0.087***(−6.21)	−0.033*(−1.67)	−0.097***(−4.62)	−0.050***(−2.87)	−0.078***(−3.08)	−0.056***(−5.18)	−0.053***(−3.01)	−0.087***(−6.41)
PLC * Beat	0.00079**(2.09)	0.0011**(1.97)	0.000091(0.11)	0.0019**(2.17)	0.00020(0.29)	0.0011(0.98)	0.00075*(1.81)	0.00061(0.85)	0.00095*(1.81)
ZScore	−0.0019***(−4.81)	−0.0019***(−6.16)	−0.0018***(−4.11)	−0.0018***(−4.74)	−0.0019***(−4.30)	−0.00091**(−2.47)	−0.0045***(−8.95)	−0.0021***(−6.37)	−0.0017***(−3.27)
ETR	0.0064(0.79)	0.0014(0.13)	0.018(1.26)	0.0064(0.43)	0.019(1.51)	−0.0097(−0.50)	−0.0048(−0.62)	0.010(0.77)	0.0028(0.29)
Big5	0.024**(2.30)	0.0080(0.77)	0.053***(3.47)	0.011(0.75)	0.038***(2.84)	0.010(0.74)	−0.0016(−0.14)	0.028**(2.28)	0.014(1.28)
NOA	−0.0022(−1.63)	−0.0030***(−3.56)	0.00087(0.55)	−0.0008(−0.69)	−0.0039***(−3.30)	−0.0026***(−2.78)	−0.0081(−1.62)	−0.0044***(−4.47)	0.0033***(2.74)
Cycle	−0.000071**(−2.25)	−0.000097***(−4.29)	−0.0000070(−0.20)	0.0000029(0.10)	−0.00009***(−2.97)	0.000069***(2.60)	−0.00053***(−15.8)	−0.000016(−0.60)	−0.00020***(−7.18)
LnMVE	−0.0074***(−4.40)	−0.0062***(−3.26)	−0.0097***(−3.07)	−0.0078**(−2.49)	−0.0084***(−3.51)	−0.0036(−0.77)	−0.019***(−10.8)	−0.0089***(−3.30)	−0.0080***(−4.47)
ROA	−0.18***(−8.94)	−0.18***(−19.9)	−0.18***(−10.8)	−0.17***(−15.7)	−0.21***(−12.3)	−0.21***(−19.0)	−0.22***(−12.0)	−0.19***(−18.2)	−0.16***(−11.6)
LnFirmAge	0.024***(6.27)	0.026***(5.53)	0.023***(3.20)	0.028***(3.84)	0.019***(3.25)	0.012(1.26)	0.011***(2.90)	0.020***(2.94)	0.025***(5.38)
ForSeg	−0.028**(−2.57)	−0.031***(−3.03)	−0.018(−1.41)	−0.010(−0.77)	−0.035***(−3.00)	−0.048***(−3.70)	−0.042***(−3.92)	−0.040***(−3.59)	−0.0094(−0.86)
RDInt	0.018***(10.6)	0.019***(18.8)	0.014***(6.80)	0.018***(13.2)	0.018***(12.4)	0.018***(15.4)	−0.46***(−13.7)	0.020***(17.2)	0.0091***(5.52)
Constant	−0.091***(−4.10)	−0.083***(−3.42)	−0.12***(−3.16)	−0.088**(−2.56)	−0.097***(−3.18)	−0.12***(−2.69)	0.21***(8.59)	−0.085***(−2.78)	−0.063**(−2.46)
Year dummy	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
No. of Obs.	20,105	13,450	6655	7468	8971	7567	12,538	10,371	9702
R square	.13	.14	.11	.15	.11	.19	.12	.17	.11

Panel B: Avoiding earnings decreases
	(A)	(B)	(C)	(D)	(E)	(F)	(G)	(H)	(I)
Variable	Full sample	W/O Mjr Cx	W/Mjr Cx	Early stage	Later stage	Low MKT share	High MKT share	Low FAI	High FAI
PLC	0.00048***(4.83)	0.00063***(4.51)	0.00028(1.37)	−0.00004(−0.19)	0.0007***(4.03)	−0.00042**(−2.08)	0.000070(0.60)	0.00049***(2.69)	0.00047***(3.45)
Increase	−0.071***(−4.00)	−0.065***(−2.96)	−0.082**(−2.43)	−0.076**(−2.00)	−0.079***(−2.95)	−0.049(−1.39)	−0.057***(−3.42)	−0.058**(−2.02)	−0.081***(−3.58)
PLC*Inc.	0.0020***(2.87)	0.0018**(1.97)	0.0023*(1.68)	0.0025(1.62)	0.0016(1.48)	0.0016(1.01)	0.0014**(2.26)	0.0019(1.53)	0.0019**(2.19)
No. of Obs.	52,955	37,506	15,449	18,579	22,095	26,459	26,496	26,409	26,421
R square	.095	.10	.085	.11	.098	.12	.13	.13	.076

Table 3 reports multivariate test results for Model (1). The dependent variables are UnexpProd in both panels. In Panels A and B, the earnings thresholds are consensus analyst forecasts and last year’s earnings, respectively. Panel B has the same set of control variables as in Panel A. Variable definitions are in Appendix.

, **, and *** indicate the statistically significant levels at 10%, 5%, and 1%, respectively.

Next, we discuss the cross-sectional tests on overproduction. The Compustat Segment data disclose names of customers that represent more than 10% of companies’ sales (major customers).¹⁴ For H3a, we create one subsample without the presence of major customers (W/O Mjr Cx) and another with (W/Mjr Cx). We rerun regression (1) in each subsample and report the results in Columns (B) and (C). In Column (B), sample firms do not have any major customers. The interaction term PLC*BEAT remains positive and significant. In Column (C), when firms have at least one major customer, the coefficient of PLC*BEAT is no longer significant. The evidence provides support for H3a—the presence of major customers weakens the relationship between product life cycle lengths and overproduction. Reputation concern likely discourages firms with major customers from overproducing, regardless of product life cycle lengths.

For H4a, following Dickinson (2011), we create a subsample of early-stage firms including firms in their introduction and growth stages, and another of later-stage firms including those in the mature, shake-out and decline stages. We rerun regression (1) in each subsample. In Column (D), we report results for early-stage firms. As predicted, the interaction term PLC*BEAT is positive and significant at a 5% level. In Column (E) of later-stage firms; however, PLC*BEAT is no longer significant. The evidence is consistent with our prediction that being in the early stages of firm life cycle exacerbates the inventory risk for short-cycle firms. In other words, being an early-stage firm strengthens the relationship between product life cycle lengths and overproduction.

Regarding H5a, we argue that being in the market-leader position also accentuates the relationship between product life cycle lengths and overproduction. We create a subsample with high market share and another with low market share based on the sample median of market share (MKTShare), where MKTShare is calculated as the percentage of a firm’s sales to total sales of its three-digit SIC industry. We rerun regression (1) in each subsample. In Column (F), for the low-market-share group, the interaction term PLC*BEAT is positive but not significant. With a low market share, firms are concerned about further eroding their competitive positions by deviating from the optimal production plan. It appears that such concerns apply to both long- and short-cycle firms, so that product life cycle lengths no longer predict firms’ overproduction choices. It is only in Column (G), when firms have a high market share, that PLC*BEAT becomes positive and significant. This provides support for H5a.

Finally, for H6, we calculate fixed-asset intensity (FAI) as fixed assets over total assets and create two subsamples based on the median of FAI. We then rerun regressions (1) in each subsample. In Column (H), among firms with low fixed-asset intensity, PLC*BEAT is insignificant. Product life cycle lengths do not affect firms’ overproduction efforts. In Column (I), among firms with high fixed-asset intensity, PLC*BEAT becomes positive and significant at the 10% level. This confirms our prediction for H6, suggesting that high fixed-asset intensity strengthens the relationship between product life cycle lengths and overproduction. Overall, results in Panel A of Table 3 provide strong support for H1 and the cross-sectional tests on overproduction (H3–H6) for the beating analyst forecasts threshold.

Next, we run the same set of tests when the earnings target is last year’s earnings and report the results in Panel B of Table 3. In this panel, the variable of interest is PLC*Increase, the interaction term of PLC and Increase. Column (A) presents the full sample results. As predicted, the coefficient of the interaction term is positive and significant at 1%. It appears that short-cycle firms also engage in overproduction less intensively than long-cycle firms to avoid earnings decreases.

In Columns (B)-(I), we present the cross-sectional test results for the avoiding earnings decreases threshold. Specifically, we report H3a results in Columns (B) and (C), H4a results in Columns (D) and (E), H5a results in Columns (F) and (G), and H6 results in Columns (H) and (I). Once again, Panel B results are generally consistent with our predictions. Together, Panels A and B of Table 3 provide support for both the main test and cross-sectional tests on overproduction. The evidence confirms that overproduction is a less desirable income-inflating choice for short-cycle firms than for long-cycle firms.

Table 4 presents results on discretionary spending cuts (including the main test H2 and the cross-sectional tests H3b–H5b). Again, we report results to beat consensus analyst forecasts in Panel A, and results to avoid earnings decreases in Panel B. In Panel A, the variable of interest is PLC*BEAT. Full sample results (for H2) are presented in Panel A, Column (A). As predicted, the coefficient of the interaction term is negative and significant at 5%. There is evidence that short-cycle firms also refrain from discretionary spending cuts more than long-cycle firms. Other controls generally have predicted signs.

Table 4.

The Impact of Product Life Cycle Lengths on Firms’ Discretionary Spending Cut.

Panel A: Beating analyst forecasts
	(A)	(B)	(C)	(D)	(E)	(F)	(G)
Variable	Full sample	W/O Mjr Cx	W/ Mjr Cx	Early stage	Later stage	Low MKT share	High MKT share
PLC	0.00068*(1.68)	0.00089(1.53)	−0.00032(−0.36)	0.00044(0.46)	0.00049(0.61)	0.00074(0.73)	0.00084(1.54)
Beat	0.098***(3.18)	0.096***(2.76)	0.091*(1.75)	0.11**(2.00)	0.077(1.64)	0.100(1.59)	0.093***(3.01)
PLC*Beat	−0.0022**(−2.26)	−0.0023*(−1.73)	−0.0010(−0.52)	−0.0028(−1.31)	−0.0014(−0.75)	−0.0034(−1.32)	−0.0016(−1.40)
ZScore	0.00094(0.90)	0.0011(1.16)	0.0013(0.95)	0.0016(1.40)	0.00069(0.48)	−0.00028(−0.25)	0.00077(0.54)
ETR	−0.011(−0.49)	−0.033(−1.34)	0.022(0.61)	0.034(0.94)	−0.066**(−2.01)	0.047(1.05)	−0.024(−1.10)
Big5	−0.037(−1.37)	−0.0036(−0.14)	−0.094**(−2.38)	−0.048(−1.23)	−0.040(−1.10)	−0.047(−1.34)	−0.032(−1.01)
NOA	−0.031***(−5.88)	−0.025***(−6.56)	−0.053***(−6.75)	−0.030***(−6.15)	−0.027***(−4.30)	−0.024***(−5.77)	−0.12***(−8.48)
Cycle	−0.000049(−0.61)	0.000013(0.21)	−0.00018*(−1.86)	−0.00013(−1.55)	0.00002(0.24)	−0.000084(−1.16)	−0.000010(−0.11)
LnMVE	−0.012***(−2.70)	−0.0077(−1.63)	−0.026***(−3.20)	−0.022***(−2.74)	−0.014**(−2.10)	−0.019(−1.48)	−0.0058(−1.18)
ROA	−0.40***(−6.14)	−0.39***(−14.2)	−0.43***(−8.66)	−0.46***(−14.4)	−0.27***(−4.81)	−0.43***(−13.4)	−0.36***(−6.92)
LnFirmAge	−0.0093(−0.93)	−0.015(−1.28)	−0.013(−0.67)	−0.0043(−0.23)	−0.014(−0.86)	−0.014(−0.60)	−0.0091(−0.84)
ForSeg	0.0084(0.27)	0.043(1.59)	−0.069*(−1.84)	0.020(0.54)	−0.026(−0.77)	0.012(0.34)	0.017(0.54)
RDInt	−0.022***(−2.64)	−0.023***(−4.05)	−0.035**(−2.48)	−0.023***(−3.05)	−0.017*(−1.80)	−0.024***(−3.84)	0.45***(4.14)
HHI	2.98***(8.32)	1.81***(3.89)	6.41***(7.42)	6.97***(6.95)	4.55***(5.76)	6.29***(5.28)	2.15***(5.20)
Constant	−0.0047(−0.072)	−0.027(−0.41)	0.060(0.60)	−0.20**(−2.12)	0.11(1.29)	−0.024(−0.20)	−0.0090(−0.12)
Year dummy	Yes	Yes	Yes	Yes	Yes	Yes	Yes
No. of Obs.	18,117	12,218	5899	6470	8068	5825	12,292
R square	.17	.17	.18	.18	.17	.24	.14

Panel B: Avoiding earnings decreases
	(A)	(B)	(C)	(D)	(E)	(F)	(G)
Variable	Full sample	W/O Mjr Cx	W/ Mjr Cx	Early stage	Later stage	Low MKT share	High MKT share
PLC	−0.00021(−0.81)	0.00028(0.77)	−0.0014***(−2.64)	−0.00039(−0.63)	−0.00086*(−1.80)	0.00033(0.63)	−0.00027(−0.80)
Increase	−0.087*(−1.66)	−0.059(−1.03)	−0.16*(−1.79)	−0.036(−0.35)	−0.18**(−2.39)	−0.13(−1.35)	−0.066(−1.35)
PLC*Increase	0.0014(0.83)	0.00038(0.17)	0.0036(1.04)	−0.0014(−0.34)	0.0049*(1.70)	−0.00083(−0.20)	0.0019(1.02)
No. of Obs.	48,371	34,359	14,012	16,319	20,240	22,379	25,992
R square	.12	.12	.14	.15	.14	.13	.13

Table 4 reports multivariate test results for Model (2). The dependent variables are UnexpDisExp in both panels. In Panels A and B, the earnings thresholds are consensus analyst forecasts and last year’s earnings, respectively. Panel (B) has the same set of control variables as in Panel (A). Variable definitions are in Appendix.

, **, and *** indicate the statistically significant levels at 10%, 5%, and 1%, respectively.

We test the moderating role of major customers for discretionary spending cuts (H3) in Columns (B) and (C), the moderating role of firm life cycle stage in Columns (D) and (E), and the moderating role of market share in Columns (F) and (G). The interaction terms are significantly negative when there is no major customer, and no longer significant when there is at least one major customer. This is again consistent with the notion that monitoring by major customers reduces firms’ incentive for RAM, regardless of product life cycle lengths. However, we did not find cross-sectional results on firm life cycle stage and market share.

In Panel B, we repeat tests for discretionary spending cuts with last year’s earnings as the earnings target. The variable of interest, PLC*Increase, is not statistically significant in the main test, Column (A). There is no evidence that short-cycle suspect firms differ from their long-cycle counterparts in beating last year’s earnings through discretionary spending cuts. Overall, the results on discretionary spending seem to be weaker than those on overproduction. One potential reason lies in our setting of manufacturing firms. It is more conducive for testing firms’ production decisions than investment decisions such as discretionary spending cuts.

Additional Test Results

Prior literature suggests that as the regulatory environment has become more stringent post-SOX, managers find it more difficult to engage in accruals manipulation. As a result, they increasingly manipulate earnings with RAM because RAM is less subject to scrutiny (Cohen et al., 2008; Eli & Cohen, 2009). In our setting, short product life cycle lengths may be effective in constraining RAM before SOX. However, as accrual manipulation becomes more costly after SOX, short-cycle firms may increasingly choose RAM despite the inventory risk and potential erosion to their competitive position. In other words, short product life cycle lengths may become a less-effective constraint on RAM in the post-SOX era. We run two additional tests to examine this possibility.

First, we run regressions (1) and (2) in the pre- and post-SOX subsamples separately and report the results in Panel A of Table 5. Columns (A) and (B) present the overproduction results for the analyst forecast threshold, and Panels (C) and (D) present overproduction results for the avoiding earnings decreases threshold. While PLC*EM is positive and significant in the pre-SOX subsamples (Columns (A) and (C)), it is no longer significant in the post-SOX tests (Columns (B) and (D)). It appears that short-cycle firms have less prevalent overproduction than long-cycle firms pre SOX, but the differences are not significant post SOX. Columns (E) to (H) present the results for discretionary spending cuts in the pre- and post-SOX era. The interaction terms do not load in any columns.

Table 5.

The Operating Constraint in the Pre- and Post-SOX Periods.

Panel A: Overproduction/discretionary spending pre- and post-SOX.
	UnexpProd and Beat		UnexpProd and Increase		UnexpDisExp and Beat		UnexpDisExp and Increase
Variable	(A) Pre-SOX	(B) Post-SOX	(C) Pre-SOX	(D) Post-SOX	(E) Pre-SOX	(F) Post-SOX	(G) Pre-SOX	(H) Post-SOX
PLC	0.00073***(4.61)	0.00089**(2.10)	0.00047***(4.81)	0.00054**(1.99)	−0.00096***(−4.18)	0.0011(0.89)	−0.00016(−1.08)	−0.0016*(−1.95)
EM	−0.066***(−7.13)	−0.074***(−2.94)	−0.052***(−3.16)	−0.096**(−2.37)	0.027**(1.99)	0.18**(2.54)	0.013(0.51)	−0.22*(−1.84)
PLC*EM	0.00069*(1.94)	0.00092(0.85)	0.0015**(2.34)	0.0025(1.51)	0.00019(0.38)	−0.0042(−1.43)	−0.0012(−1.22)	0.0046(0.92)
Constant	−0.091***(−5.57)	0.21***(5.72)	−0.15***(−15.3)	0.15***(6.10)	0.13***(5.25)	−0.34***(−3.02)	−0.088***(−5.53)	−0.51***(−6.30)
Year dummy	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
# of Obs.	11847	8258	33510	19445	10905	7212	31083	17288
R square	.14	.13	.11	.10	.11	.18	.1092	.13

Panel B: The impact of SOX on overproduction within short- vs. long-cycle firms
	Short-cycle firms		Long-cycle firms
Variable	(A) Beat	(B) Increase	(C) Beat	(D) Increase
SOX	0.33*(1.82)	0.18(0.75)	−0.069(−1.09)	−0.0083(−0.088)
ZScore	−0.0027*(−1.81)	−0.00054(−0.49)	−0.0024***(−3.56)	−0.0025*(−1.90)
ETR	−0.060(−1.61)	−0.25**(−2.33)	0.0095(0.47)	−0.017(−0.23)
Big5	0.053(1.24)	0.051(1.02)	0.0068(0.27)	−0.034(−0.84)
NOA	−0.0056(−0.82)	−0.0030(−0.55)	0.017***(4.01)	0.013(1.60)
Cycle	−0.000040(−0.40)	0.00000067(0.0043)	−0.00044***(−4.32)	−0.00019(−0.94)
LnMVE	−0.023***(−3.63)	−0.029***(−3.30)	−0.015***(−3.27)	−0.0019(−0.28)
ROA	−0.25**(−2.06)	−0.0080(−0.10)	−0.19*(−1.71)	−0.19(−1.43)
LnFirmAge	0.053***(3.71)	0.055**(2.50)	0.043***(4.86)	0.0054(0.37)
ForSeg	−0.029(−0.82)	−0.065(−1.17)	−0.013(−0.44)	0.029(0.63)
RDInt	0.024***(2.63)	0.032***(3.13)	0.0070(0.84)	0.025***(2.64)
Constant	−0.12(−1.36)	−0.084(−0.76)	−0.037(−0.67)	−0.038(−0.39)
Year dummy	Yes	Yes	Yes	Yes
No. of Obs.	1972	720	1726	628
R square	.17	.16	.12	.22

Table 5 reports results on the constraining role of short product cycle lengths before and after SOX. Panel (A) reports Models (1) and (2) results for the pre- and post-SOX subsamples, respectively. Panel B reports results of Model (3). The dependent variables are UnexpProd in all columns. Samples in Columns (A) and (C) include short-cycle and long-cycle firm years that beat consensus analyst forecast by no more than 1 cent, respectively. Samples in Columns (B) and (D) include short-cycle and long-cycle firm-years with changes in earnings from last year falling in the bin immediately to the right of zero, respectively.

, **, and *** indicate the statistically significant levels at 10%, 5%, and 1%, respectively.

Panel A suggests that while short product life cycle lengths are effective in constraining overproduction pre-SOX, short- and long-cycle firms are comparable in overproduction in the post-SOX era. Two reasons could be responsible for the results. Either long-cycle firms reduce overproduction or short-cycle firms increase their engagement in overproduction post SOX. To find out which is the case, we run the following regression model within the subsample of short-cycle (long-cycle) firms that narrowly beat earnings benchmarks. A firm is defined as narrowly beating benchmarks if its actual EPS is equal to or greater than consensus forecast by no more than 1 cent, or if its change in net income from last year to this year falls in the bin immediately to the right of zero.

\begin{matrix} UnexpProd = β_{0} + β_{1} SOX + β_{2} ZScore + β_{3} ETR + β_{4} Big 5 + β_{5} NOA + \\ β_{6} Cycle + β_{7} LnMVE + β_{8} ROA + β_{9} LnFirmAge + β_{10} ForSeg + \\ β_{11} RDInt + Year Dummies + ε . \end{matrix}

(3)

SOX is coded as 1 if the firm-year is after 2002, and 0 otherwise. Column (A) of Panel B presents results within short-cycle firms that narrowly beat analyst forecasts. The dummy variable SOX is positive and significant. This suggests that short-cycle firms have increased their levels of overproduction in the post-SOX period. Column (B) presents results within short-cycle firms that narrowly beat last year’s earnings. SOX is positive but insignificant. Next, we regress overproduction against SOX in subsamples of long-cycle suspect firms. The variable SOX is not statistically significant whether the threshold is analyst forecasts (Column (C)) or last year’s earnings (Column (D)). It appears that long-cycle firms engage in similar levels of overproduction before and after SOX. From results in Panel B, it appears that the weaker post-SOX results in Panel A stem from short-cycle firms’ increasing engagement in overproduction. In other words, short product life cycle lengths become less effective as a constraint on RAM after SOX. The evidence highlights the mixed effects of SOX on firms’ earnings management.¹⁵

In the main tests, we find evidence that short-cycle firms cut discretionary spending to beat analyst forecasts to a lesser extent than long-cycle firms. Vorst (2016) suggests that firms often reverse an abnormal spending cut in the year after the cut. Next, we examine whether short-cycle firms that cut discretionary spending to meet earnings targets are more likely to reverse the cut than their long-cycle counterparts. Specifically, we run the following regression within firms that cut discretionary spending in the current year.

\begin{matrix} UnexpDisEx p_{Lead} = β_{0} + β_{1} PLC + β_{2} EM + β_{3} PLC * EM + β_{4} ZScore + β_{5} ETR + \\ β_{6} Big 5 + β_{7} NOA + β_{8} Cycle + β_{9} LnMVE + β_{10} ROA + β_{11} LnFirmAge + \\ β_{12} ForSeg + β_{13} RDInt + β_{14} HHI + Year Dummies + ε, \end{matrix}

(4)

where UnexpDisExp_Lead is the unexpected discretionary spending in the subsequent year. Since opportunistic spending cuts are more costly for short-cycle firms, short-cycle firms should have more incentives to reverse the spending cut to reduce the potential negative economic impacts in the subsequent year. We predict a negative coefficient for PLC*Beat (PLC*Increase). The results are presented in Table 6.

Table 6.

The Impact of Product Life Cycle on Reversal of Discretionary Spending Cut.

Variable	(A) Beat	(B) Increase
PLC	0.0030*** (5.08)	0.0015*** (4.19)
EM	0.065 (1.48)	−0.090 (−1.34)
PLC*EM	−0.0032** (−2.25)	0.0021 (0.96)
Z score	0.0050*** (4.00)	0.0023** (2.21)
ETR	−0.031 (−1.05)	−0.035* (−1.84)
Big5	0.0041 (0.11)	0.036** (1.96)
NOA	−0.0079 (−0.80)	−0.0069 (−1.34)
Cycle	−0.00018 (−1.27)	−0.000070 (−0.81)
LnMVE	−0.015*** (−2.71)	0.00024 (0.071)
ROA	−0.0090 (−0.10)	−0.071 (−1.35)
LnFirmAge	0.0079 (0.61)	−0.0050 (−0.58)
ForSeg	0.072* (1.65)	0.038 (1.62)
RDInt	−0.022 (−1.57)	−0.024* (−1.89)
HHI	5.80*** (13.2)	3.97*** (15.3)
Constant	−0.42*** (−5.67)	−0.33*** (−7.39)
Year dummy	Yes	Yes
No. of Obs.	9464	24,588
R square	.18	.13

Table 6 reports reversal test results for Model (4). The dependent variables in both columns are UnexpDisExp_Lead, that is, unexpected discretionary spending in year t+ 1. The sample includes all firm-years that cut discretionary spending in year t.

, **, and *** indicate the statistically significant levels at 10%, 5%, and 1% respectively.

In Table 6, PLC*EM is significantly negative when the earnings threshold is consensus analyst forecasts, but PLC*EM is statistically insignificant when the threshold is last year’s earnings. Together with the main test results, the evidence suggests the following: (a) short-cycle firms are less likely to cut discretionary spending to meet analyst forecasts than long-cycle firms. (b) Among firms that choose to cut discretionary spending to meet analyst forecasts, short-cycle firms are more likely to reverse the spending cut in the next period. Both results are consistent with the expectation that discretionary spending cuts are more costly for short-cycle firms than for long-cycle firms.

One earnings management method related to overproduction is LIFO liquidation. LIFO liquidation occurs when a LIFO company depletes older and cheaper inventory. By doing so, the current-period cost of goods sold would be lower and net income would be higher. Prior literature suggests that managers use LIFO liquidation to boost earnings (e.g., Dhaliwal et al., 1994). Since some short-cycle firms use LIFO as an inventory costing method and can use LIFO liquidation to boost earnings, our H1 results can hold either because long-cycle firms indeed overproduce more than short-cycle ones in suspect years, or because short-cycle LIFO firms liquidate old LIFO layers to a much larger extent than long-cycle firms in suspect years. To ensure that our H1 results are not driven by more intensive LIFO liquidation by short-cycle firms, we run the following test within all firms that use LIFO as one of their inventory costing methods:¹⁶

\begin{matrix} LIFO Lqdt = β_{0} + β_{1} PLC + β_{2} EM + β_{3} PLC * EM + β_{4} ZScore + β_{5} ETR + \\ β_{6} Big 5 + β_{7} NOA + β_{8} Cycle + β_{9} LnMVE + β_{10} ROA + β_{11} LnFirmAge + \\ β_{12} ForSeg + β_{13} RDInt + β_{14} ChgSales + β_{15} LIFR + Year Dummies + ε . \end{matrix}

(5)

We identify LIFO liquidation instances following Dhaliwal et al. (1994).¹⁷ The results are presented in Table 7. Column (A) presents the results to beat consensus analyst forecasts. PLC*EM is positive and significant at 5%. It appears that short-cycle LIFO firms also engage in LIFO liquidation to a lesser extent than long-cycle firms do. In Column (B), the earnings target is last year’s earnings and the coefficient of PLC*EM is positive but insignificant. Untabulated results suggest that on average, short-cycle firms carry lower levels of LIFO reserves than long-cycle firms ($5.38 million versus $10.18 million). This limits the usefulness of LIFO liquidation as an income-increasing choice for short-cycle firms. This robustness check boosts our confidence that H1 results are indeed due to excessive production by long-cycle suspect firms.

Table 7.

The Impact of Product Life Cycle Lengths on Firms’ LIFO Liquidation.

Variable	(A) Beat	(B) Increase
PLC	−0.022** (−2.38)	−0.0091* (−1.82)
EM	−1.55** (−2.19)	−0.52 (−0.60)
PLC*EM	0.050** (2.44)	0.019 (0.75)
ZScore	−0.20** (−2.49)	−0.13*** (−2.71)
ETR	−0.0066 (−0.023)	0.10 (0.48)
Big5	−0.52 (−0.94)	0.36 (1.34)
NOA	−1.85*** (−3.97)	−1.43*** (−4.58)
Cycle	0.0023 (1.28)	0.0012 (0.93)
LnMVE	−0.13 (−1.31)	−0.13** (−2.29)
ROA	−0.48 (−0.23)	−0.91 (−1.34)
LnFirmAge	0.29** (1.99)	0.22** (2.12)
ForSeg	−0.96*** (−2.89)	−0.48** (−2.15)
RDInt	−11.3* (−1.81)	−15.0*** (−3.06)
ChgSales	−0.79* (−1.72)	−0.42 (−1.49)
LIFR	0.0040** (2.23)	0.0027** (2.04)
Constant	0.85 (0.90)	−0.93 (−1.46)
Year dummy	Yes	Yes
No. of Obs.	1176	2605
Pseudo R square	.18	.13

Table 7 reports multivariate test results for Model (5) within firm years that use LIFO as one of their inventory costing methods. The dependent variable is LIFO Lqdt, coded as 1 if the liquidation occurred in the current year and was not reported as part of discontinued operations, and 0 otherwise.

, **, and *** indicate the statistically significant levels at 10%, 5%, and 1%, respectively.

In the main test, we measure the product life cycle lengths with PLC, a rank variable of the product life cycle lengths in years (T). As a robustness check, we replace PLC with T in regressions (1) and (2) and present the results in Table 8. In Columns (A) and (B), we present the results on overproduction. The interaction term T*EM is positive and significant for both thresholds. Columns (C) and (D) present the results for discretionary spending regression for the two thresholds. T*EM is significantly negative in the analyst forecasts test but insignificant when the threshold is last year’s earnings. The above results are qualitatively the same as those in Tables 3 and 4. Overall, our main test results are robust to using the alternative measure for the product life cycle lengths.

Table 8.

Product Life Cycle Lengths With Number of Years in PLC.

	UnexpProd		UnexpDisExp
	(A)	(B)	(C)	(D)
Variable	Beat	Increase	Beat	Increase
T	0.018*** (4.69)	0.013*** (5.24)	0.0050 (0.51)	−0.015** (−2.48)
EM	−0.27*** (−2.82)	−0.41*** (−2.58)	0.50** (2.21)	−0.38 (−0.96)
T*EM	0.024** (2.31)	0.040** (2.43)	−0.048** (−2.03)	0.035 (0.83)
Z score	−0.0019*** (−4.86)	0.000079 (0.31)	0.00093 (0.88)	0.0020** (2.46)
ETR	0.0067 (0.83)	−0.011** (−2.08)	−0.010 (−0.44)	0.0067 (0.43)
Big5	0.024** (2.27)	0.024*** (4.70)	−0.037 (−1.37)	0.033** (2.45)
NOA	−0.0022* (−1.65)	0.00017 (0.21)	−0.031*** (−5.88)	−0.024*** (−7.54)
Cycle	−0.000071** (−2.22)	0.000032* (1.77)	−0.000047 (−0.59)	0.00013** (2.38)
LnMVE	−0.0075*** (−4.45)	−0.0065*** (−7.16)	−0.013*** (−2.73)	0.0095*** (3.77)
ROA	−0.18*** (−8.93)	−0.094*** (−11.9)	−0.40*** (−6.13)	−0.38*** (−12.6)
LnFirmAge	0.024*** (6.27)	0.025*** (9.87)	−0.0078 (−0.78)	−0.013** (−1.97)
ForSeg	−0.027** (−2.45)	−0.024*** (−3.92)	0.0084 (0.27)	0.036** (2.07)
RDInt	0.018*** (10.5)	0.021*** (18.5)	−0.022*** (−2.65)	0.0030 (0.37)
HHI			3.02*** (8.29)	1.58*** (7.33)
Constant	−0.24*** (−5.69)	−0.24*** (−9.12)	−0.041 (−0.36)	−0.069 (−1.00)
Year dummy	Yes	Yes	Yes	Yes
No. of Obs.	20,105	52,955	18,117	48,371
R square	.13	.095	.17	.12

Table 8 reports robustness test results for Model (1) and (2). Specifically, we replace PLC rank with product life cycle lengths in years (T) to capture product life cycle lengths.

, **, and *** indicate the statistically significant levels at 10%, 5%, and 1%, respectively.

In the main test, we control firm life cycle stage using the natural log of firm age. In an unreported test, we use alternative measures to capture this construct. Following Dickinson (2011), we create four dummy variables to capture the growth, mature, shake-out, and decline stages based on firms’ cash flow patterns. We rerun the main tests replacing firm age with these dummy variables. The results remain qualitatively the same.

Since some products are positively correlated with macro performance, while others are Giffen goods, we examine how macroeconomic conditions affect product demand and hence our results. We expect that during contractionary periods, most firms are experiencing lower productivity and poorer financial performance and do not have the same incentives to manage earnings. Following Klein and Marquardt (2006), we measure the economic cycle with the NBER definition of expansionary and contractionary periods. We create two subsamples for periods of expansion versus contraction, and rerun tests within each subsample. In the unreported test, we find predicted results in the expansion subsample, but not in the contraction subsample. It is plausible that during contraction, both long- and short-cycle firms focus on surviving the economic hardship rather than engaging in RAM.¹⁸

Capacity constraints may affect firms’ production decisions. To ensure that capacity utilization does not drive our test results, we create a variable Capacity. It is calculated as the annual mean capacity utilization of manufacturing industries (as reported in the Federal Reserve’s “Industrial Production and Capacity Utilization.”). In an unreported test, we control Capacity in the main tests and our inferences remain the same.¹⁹

Conclusion

This article examines the role of short product life cycle lengths in constraining RAM. Product life cycle lengths are an important concept in the operations management literature. Short-cycle firms have to predict and closely follow customers’ changing tastes and preferences. Rapidly advancing technology also increases pressure on short-cycle firms to develop new products that incorporate the newest technology. Uncertainty and rapid changes in demand increase the inventory risk and cost of opportunist spending cuts. In other words, short-cycle firms’ inventory management and SG&A planning likely follow very different principles and processes than those of long-cycle firms. We argue that short product life cycle lengths limit firms’ engagement in RAM such as overproduction and discretionary spending cuts.

As predicted, we document that short-cycle firms achieve earnings targets through RAM to a lesser extent than their long-cycle counterparts do. In an additional test, we find that short product life cycle lengths effectively constrain overproduction before SOX. Post SOX, short-cycle firms increased their engagement in overproduction to such an extent that there is no longer a significant difference between short- and long-cycle firms. This finding highlights the unintended consequences of SOX. As RAM is less subject to regulatory oversight, its popularity soared post-SOX. Operating constraints such as short product life cycle lengths become less effective. Once again, our finding calls attention to the notion that SOX does not necessarily enhance the decision usefulness of earnings (Dechow et al., 2010).

Our main contribution lies in the identification of an operating constraint on RAM (i.e., short product life cycle lengths). With the increasing popularity and potential negative impact of RAM, researchers and investors have more incentives to detect and constrain RAM. Constraints can stem from a firm’s operating characteristics or corporate governance. In the literature, researchers have identified several corporate governance factors and litigation threats that limit managers’ abilities to engage in RAM (e.g., Chen et al., 2015; Huang et al., 2020). One of the takeaways of this article is that the detection of RAM is more likely in long-cycle firms, or in the post-SOX period in short-cycle firms.

In the operations management literature, researchers often use anecdotal evidence or theoretical models to study the optimal production and inventory levels in short-cycle firms. Our paper provides empirical evidence on how such firms make operating decisions. Future studies could build on theories from the operations research and expand the empirical exploration for factors that influence firms’“normal” levels of production and discretionary spending. For example, incorporating the dimension of product life cycle lengths in the RAM expectation models may improve the model fitting. In addition, firms concerned with stock-outs and lost sales are more likely to keep large safety stock. According to the operations research literature, stock-outs often lead to order backlogs rather than lost sales for manufacturers that sell to wholesalers. If so, manufacturers that sell to wholesalers face lower stock-out costs and likely have a lower level of optimal inventory than manufacturers that sell products to retailers. Future studies could test this prediction and further improve the RAM expectation models. Similarly, borrowing from theories in operations research could also enhance our understanding of factors that go into the planning of discretionary spending, and help us distinguish normal spending from opportunistic spending. Such an understanding would ultimately improve security valuation related to unexpected changes in inventory, R&D, and SG&A costs.

Footnotes

Appendix

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Professor Meng Yan thanks the Gabelli School of Business at Fordham University for generous research support.

ORCID iD

Crystal Xu

Notes

Author Biographies

Professor Crystal Xu is an Assistant Professor at Manhattan College. She joined the O’Malley School Of Business, Accounting, Business Analytics, Cis & Law Department in 2015.

Professor Meng Yan is an Associate Professor in the Accounting and Taxation area at Gabelli School of Business. She has been teaching at Fordham since 2007.

References

Abedi

V. S.

Berman

Krass

(2014). Supporting new product or service introductions: Location, marketing, and word of mouth. Operations Research, 62, 994–1013.

Anderson

M. C.

Banker

R. D.

Janakiraman

S. N.

(2003). Are selling, general, and administrative costs “sticky”? Journal of Accounting Research, 41, 47–63.

Bauer

A. M.

Henderson

Lynch

D. P.

(2018). Supplier internal control quality and the duration of customer-supplier relationships. The Accounting Review, 93, 59–82.

Bilir

L. K.

(2014). Patent laws, product life-cycle lengths, and multinational activity. American Economic Review, 104, 1979–2013.

Bushee

(1998). The inﬂuence of institutional investors on myopic R&D investment behavior. The Accounting Review, 73, 305–333.

Chen

J. Z.

Cussatt

Gunny

K. A.

(2020). When are outside directors more effective monitors? Evidence from real activities manipulation. Journal of Accounting, Auditing & Finance, 35, 26–52.

Chen

Cheng

A. K.

Wang

(2015). CEO contractual protection and managerial short-termism. The Accounting Review, 90, 1871–1906.

Cheng

Lee

Shevlin

(2016). Internal governance and real earnings management. The Accounting Review, 91, 1051–1085.

Cohen

D. A.

Dey

Lys

T. Z.

(2008). Real and accrual-based earnings management in the pre- and post-Sarbanes-Oxley periods. The Accounting Review, 83, 757–787.

10.

Cohen

D. A.

Zarowin

(2010). Accrual-based and real earnings management activities around seasoned equity offerings. Journal of Accounting and Economics, 50, 2–19.

11.

Dechow

Schrand

(2010). Understanding earnings quality: A review of the proxies, their determinants and their consequences. Journal of Accounting and Economics, 50, 344–401.

12.

Dhaliwal

D. S.

Frankel

Trezevant

(1994). The taxable and book income motivations for a LIFO layer liquidation. Journal of Accounting Research, 32, 278–289.

13.

Dickinson

(2011). Cash flow patterns as a proxy for firm life cycle. The Accounting Review, 86, 1969–1994.

14.

Eli

Cohen

(2009). The “numbers game” in the pre- and post-Sarbanes-Oxley eras. Journal of Accounting, Auditing & Finance, 24, 505–534.

15.

Gaudi

F. C.

Johnson

E. S.

Myers

J. N.

Myers

L. A.

(2019). Accounting for inventory costs and real earnings management behavior [Working paper]. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3847331

16.

Graham

J. R.

Harvey

C. R.

Rajgopal

(2005). The economic implications of corporate financial reporting. Journal of Accounting and Economics, 40, 3–73.

17.

Gupta

Pevzner

Seethamraju

(2010). The implications of absorption cost accounting and production decisions for future firm performance and valuation. Contemporary Accounting Research, 27, 889–922.

18.

Hauser

J. R.

Shugan

(2008). Defensive marketing strategies. Marketing Science, 27, 85–87.

19.

Hossain

Mitra

Rezaee

Sarath

(2011). Corporate governance and earnings management in the pre- and post-Sarbanes-Oxley act regime: Evidence from implicated option backdating firms. Journal of Accounting, Auditing & Finance, 26, 279–315.

20.

Huang

Roychowdhury

Sletten

(2020). Does litigation deter or encourage real earnings management? The Accounting Review, 95, 251–278.

21.

Hui

K. W.

Klasa

Yeung

P. E.

(2012). Corporate suppliers and customers and accounting conservatism. Journal of Accounting and Economics, 53, 115–135.

22.

Klein

Marquardt

(2006). Fundamentals of accounting losses. The Accounting Review, 81, 179–206.

23.

Kurawarwala

A. A.

Matsuo

(1996). Forecasting and inventory management of short life-cycle products. Operations Research, 44, 131–150.

24.

Martinez-de-Albeniz

Simchi-Levi

(2009). Competition in the supply option market. Operations Research, 57, 1082–1097.

25.

Payne

Robb

(2000). Earnings management: The effect of ex ante earnings expectations. Journal of Accounting, Auditing & Finance, 15, 371–392.

26.

Roychowdhury

(2006). Earnings management through real activities manipulation. Journal of Accounting and Economics, 42, 335–370.

27.

Srivastava

(2019). Improving the measure of real earnings management. Review of Accounting Studies, 24, 1277–1316.

28.

Stout

Hanson

(1989). Controlling inventory obsolescence. Purchasing World, 33, 36–38.

29.

Vorst

(2016). Real earnings management and long-term operating performance: The role of reversals in discretionary spending cuts. The Accounting Review, 91, 1219–1256.

30.

Zang

A. Y.

(2012). Evidence on the trade-off between real activities manipulation and accrual-based earnings management. The Accounting Review, 87, 675–703.