Insider Trading Signals Across Industries: Evidence from Technology, Utilities, and Banking

Abstract

This paper examines how the predictive content of insider trading varies across industries. Using U.S. insider transaction data from 2005 to 2025 and firm-month level measures of insider trading and forward returns, we compare technology, banking, and utility firms within a unified framework. The results show that insider purchases in banking firms contain the strongest information about future returns, while the signal is substantially weaker in technology firms and moderate in utilities. We also document a clear asymmetry between buying and selling. Insider purchases are more informative than sales, while sales reflect more heterogeneous motives and are therefore harder to interpret. This buy–sell gap varies across industries and is most pronounced in banking and utilities. Finally, we compare insider-trading informativeness before and after the 2022 amendments to Rule 10b5-1. The results show that sell-side informativeness appears weaker in the post-2023 period, while the predictive content of purchases remains largely unchanged. This evidence is descriptive and does not imply a causal effect of the reform. Overall, the findings highlight the importance of industry-specific information environments and regulatory conditions in shaping the relation between insider trading and future stock returns.

1. Introduction

Insider trading is widely viewed as a key channel through which private information enters financial markets. A large body of evidence shows that insider trades can predict future stock returns, suggesting that prices do not immediately reflect private information (Jaffe, 1974; Seyhun, 1986; Lakonishok & Lee, 2001). These findings imply that insider trading remains informative, even in markets with strong disclosure requirements and regulatory oversight.

At the same time, firms operate in very different information environments across industries. Technology firms rely heavily on research and development and intangible assets, which are difficult for outside investors to evaluate, giving insiders a stronger informational advantage (Aboody & Lev, 2000). Banks, in contrast, operate under strict regulatory supervision, in which disclosure requirements and monitoring limit both the frequency and the nature of insider trading (Belkhir, 2005). Utility firms are characterized by stable operations, regulated pricing, and relatively transparent reporting environments, which may make insider transactions easier to interpret (Bushman et al., 2004). These differences suggest that the predictive content of insider trading may vary systematically across industries.

In addition to industry differences, regulatory changes may also shape how insider trading signals are formed and interpreted. Recent amendments to SEC Rule 10b5-1 reflect growing concern over the potential misuse of prearranged trading plans and aim to strengthen oversight of insider transactions. Together, variation in industry information environments and changes in regulatory conditions raise a central question: does the informativeness of insider trading differ across industries, and how does it evolve over time?

Prior research has established several important features of insider trading. First, insider trades—particularly purchases—are often associated with future stock performance. Second, the information content of insider trading depends on firm-level characteristics such as information asymmetry and disclosure quality. Third, there is a well-documented asymmetry between insider purchases and sales, with purchases generally viewed as more informative. However, these dimensions are rarely examined jointly within a unified empirical framework. As a result, there is limited evidence on whether insider trading signals differ systematically across industries when differences in information environments, buy–sell asymmetry, and regulatory conditions are considered together.

To address this gap, this paper examines insider-trading informativeness across industries within a unified framework that jointly considers industry differences, buy–sell asymmetry, and changes in the regulatory environment. Understanding how these factors interact is important for evaluating how private information is incorporated into stock prices in modern financial markets.

The empirical results show several patterns that differ across industries. Insider purchases are the clearest example. These trades are more strongly related to future returns for banking firms, while the effect is much weaker for technology companies and only moderately visible for utilities. The difference is consistent with the information environment in each sector: banks operate with heavy supervision and transparent reporting, so trades made by insiders are more likely to reflect genuine private assessments of fundamentals. Technology firms, in contrast, release frequent news and analyst updates, making it harder to separate private information from the constant flow of public signals. Utilities fall somewhere in between, with stable operations that allow insider purchases to matter, though not to the same extent as in the banking sector.

The analysis also confirms a well-known asymmetry between purchases and sales. Insider buying tends to convey clearer information than selling, a pattern that aligns with earlier studies showing that sales are often associated with liquidity needs or portfolio adjustments, but they may also include trades based on private information, making their overall signal more difficult to interpret. In this dataset, sell-side activity shows some association with future returns, but the effect is smaller and far less consistent than that of the purchase signal.

A third result is that the buy–sell asymmetry itself varies across industries. In banking and utility firms, the gap between the informativeness of purchases and sales is more pronounced, while technology firms show much weaker differentiation. This suggests that industry structure influences not only the strength of insider signals but also the relative meaning of buy and sell transactions. The evidence, therefore, provides only partial support for the third hypothesis, as the asymmetry is clearly present but not uniform across sectors.

Finally, the study examines whether changes in trading rules alter these patterns. In the post-2023 period, insider sales appear less informative, while the predictive content of purchases remains essentially unchanged. The results are consistent with the possibility that the post-amendment regulatory environment is associated with weaker sell-side informativeness, while buy-side informativeness appears largely unchanged.

The first contribution of this study is to document apparent cross-industry differences in the informativeness of insider trades. Prior research has shown that insiders can earn abnormal returns; however, most studies treat insiders as a single group, without distinguishing how information flows across sectors. The evidence in this paper shows that insider purchases do not generate the same signal in technology, banking, and utility firms. Instead, the strength of the signal aligns closely with each industry’s information environment. Banking insiders appear to trade in settings where private information is more concentrated and less frequently revealed through public channels, while technology insiders operate in an environment with frequent disclosures, active analyst coverage, and fast information diffusion. Utilities fall between these two extremes. By comparing these sectors directly, the study provides a clearer picture of how industry structure shapes the value of insider trading signals. This area has received limited attention in the existing literature. By placing industry heterogeneity at the center of the analysis, this paper extends the insider-trading literature beyond aggregate market-level evidence.

A second contribution of this study is to show that the commonly documented buy–sell asymmetry is not uniform across industries. Prior work generally finds that insider purchases are more informative than sales, but most studies treat this pattern as a broad market regularity. The results here suggest that the asymmetry depends on each industry’s disclosure practices, regulatory pressure, and the way information is produced and released. In sectors such as banking and utilities, where oversight is stronger and public disclosures are more standardized, insider selling appears more constrained, making purchases relatively more informative. In contrast, technology firms operate in environments with frequent news events, higher uncertainty, and compensation structures that often involve equity-based incentives, which can blur the distinction between information-driven trades and routine portfolio adjustments. As a result, the buy–sell gap becomes much weaker. This evidence implies that the usual asymmetry is shaped by industry-specific conditions rather than a universal behavioral rule.

The third contribution of this study is to document how the informativeness of insider trades differs before and after the 2022 amendments to Rule 10b5-1. While insider trading research has focused mainly on long-standing patterns in buying and selling activity, less is known about how these patterns evolve following changes in the regulatory environment. The evidence in this paper shows a noticeable decline in the predictive content of sell-side trades in the post-2023 period, while the information contained in purchases remains largely unchanged.

This pattern is consistent with the possibility that changes in the regulatory environment are associated with differences in sell-side trading behavior. At the same time, because the current data do not distinguish between plan-based and non-plan trades, the evidence should be interpreted as descriptive rather than causal. More broadly, the results provide early evidence on how insider-trading signals evolve following regulatory changes, without isolating the specific mechanisms through which such changes operate.

By examining these issues jointly within a single framework, this study provides a more comprehensive view of how private information is incorporated into stock prices across different regulatory and informational environments.

The remainder of the paper is organized as follows. Section 2 reviews the related literature and develops hypotheses. Section 3 describes the data sources, variable construction, and empirical design. Section 4 presents the main empirical results and additional analyses. Section 5 concludes.

2. Literature Review and Hypotheses Development

2.1. Insider Trading and Market Efficiency

The Efficient Market Hypothesis (EMH) suggests that stock prices fully reflect all available information, implying that no one, including insiders, can consistently achieve above-average returns by trading on private information. Despite this, early research challenged the assumption by demonstrating that insiders could consistently generate abnormal profits by trading on information not yet publicly disclosed. One of the first studies to highlight this was Jaffe’s (1974) work, which found that insiders—individuals with access to non-public information—could profit significantly by trading before the market adjusted. This finding casts doubt on the idea that markets are always perfectly efficient.

Further evidence, as presented by Seyhun (1986), reinforces this view. His study shows that insiders not only earned abnormal returns, but outsiders who mimicked insider trading strategies also enjoyed profits. This suggested that the information insiders traded on was not immediately reflected in market prices, which would otherwise make it difficult for outsiders to replicate insiders’ profits. However, Seyhun’s study also noted that these profits were not infinite and that transaction costs constrained outsiders’ ability to profit from insider information.

In response to concerns that insiders’ profitability might have been overstated, Rozeff and Zaman (1988) refined earlier findings. They pointed out that once transaction costs were taken into account, the abnormal returns earned by outsiders attempting to mimic insider trades were significantly reduced. This suggested that, while markets were not perfectly efficient, they did exhibit some level of efficiency. However, this raised an important question: under what conditions do insiders continue to earn abnormal profits?

Building on this, Lakonishok and Lee (2001) further expanded the discussion by showing that insider trading can indeed signal future stock returns. However, they found that the predictive power of insider trading varies depending on the type of trade being made. Insider purchases, especially in smaller firms, were better indicators of future stock performance, while insider sales did not provide the same predictive value. This distinction is significant because it suggests that the market tends to react more strongly to insider expectations of positive developments than to signals of adverse outcomes.

Cohen et al. (2012) took this a step further by differentiating between “opportunistic” and “routine” insider trades. They argued that only opportunistic trades—those driven by significant private information about the company’s future—are likely to result in abnormal returns. Routine trades, however, which are part of an insider’s regular investment strategy, do not have the same predictive power. This distinction helps clarify the role of insider trading in the market, emphasizing that not all insider transactions are equally informative. What matters is not just the trade itself, but the motivation behind it and the quality of the private information that drives the decision.

Of course, part of the cross-sectional variation in returns can be explained by standard risk factors such as size, value, liquidity, or volatility (Fama & French, 1993, 2015; Amihud, 2002; Huang et al., 2025). What the insider-trading literature shows is that, even after accounting for these more mechanical return drivers, trades made by corporate insiders still appear to contain incremental information about future performance.

These studies, both early and more recent, have shaped our understanding of insider trading and market efficiency. While they challenge the notion of perfectly efficient markets, they also highlight the importance of considering the nature of insider trades. Market conditions and the characteristics of these trades are critical when evaluating their predictive power for stock returns. This insight is especially relevant when insider trading is compared across industries with different information environments.

2.2. Predictive Power and Long-Term Performance

Generally, prior research on insider trading broadly examined short-term price reactions, focusing on how the market responds when insiders buy or sell their company’s stock. Over time, however, scholars began to ask whether the informational edge insiders hold might also influence longer-term performance. Pettit and Venkatesh (1995) were among the first to study this systematically. They found that firms experiencing substantial insider buying tended to outperform the market for up to three years following the insider buying. Their results suggest that insiders trade not only on short-term signals but also on information about a firm’s underlying fundamentals and future growth.

Evidence from other markets supports this pattern. Del Brio et al. (2002) investigated insider trading in Spain, and showed that insiders consistently earn excess profits by acting on private information, whereas outside investors who attempt to replicate these trades were unsuccessful. This outcome suggests that insiders possess insights into a company’s prospects that are difficult for public investors to discern. Likewise, Friederich et al. (2002), using data from the United Kingdom, find clear short-term abnormal returns around directors’ trades. They observed that trades of moderate size often conveyed the strongest information, possibly because informed insiders prefer to hide their activity within normal trading volumes.

Jeng et al. (1999) provided complementary evidence by constructing portfolios of insider purchases and sales and tracking their performance over a six-month period. Their purchase-based portfolios earned significant positive abnormal returns, reinforcing the view that insider buying can reveal valuable information about future performance. Taken together, these studies show that insider transactions—particularly purchases—often precede periods of above-average returns, a pattern observed across different countries and time horizons.

Overall, this body of work suggests that insiders act on knowledge that reaches beyond temporary mispricing, reflecting deeper awareness of their firms’ long-term outlooks. Nevertheless, the predictive power of this approach is far from uniform. It depends on the firm’s characteristics, the industry context, and the broader institutional environment. These differences naturally lead to a central question: why do some insider trades convey stronger signals than others? The following section addresses this issue by examining how information asymmetry and disclosure environments influence the informativeness of insider trading.

2.3. Information Asymmetry and Informational Environment

The profitability of insider trading is not uniform across firms or industries, and a key explanation lies in differences in their information environments. Firms vary in the amount of private information they hold and in how effectively this information is transmitted to the market. When public disclosures are limited or analyst coverage is sparse, insiders enjoy a greater informational advantage, allowing their trades to become more predictive of future performance. Recent evidence further shows that the profitability of insider trading is closely tied to firms’ information environments and the extent of informational frictions (e.g., Ravina & Sapienza, 2010; Beneish et al., 2015), reinforcing the role of information asymmetry as a central driver of insider trading gains.

Aboody and Lev (2000) provide one of the clearest illustrations of this mechanism by examining R&D-intensive firms. They find that investments in research and development, which are difficult for outsiders to evaluate, generate substantial information asymmetry between insiders and the market. As a result, insiders at such firms earn significantly higher abnormal returns, suggesting that intangible assets amplify insiders’ informational edge due to weak and delayed disclosure. Huddart and Ke (2007) further confirm that the strength of insider–return relationships depends on firm-level opacity: insider trades are more informative in firms where external information channels are less effective.

Theoretical work offers similar insights. Easley and O’Hara (2004) argue that the mix of public and private information available to investors affects expected returns, as uninformed investors demand compensation for the risk associated with information asymmetry. Lambert et al. (2007) also demonstrate that higher disclosure quality reduces a firm’s cost of capital by enhancing the accuracy of investors’ beliefs. These findings collectively suggest that the transparency of a firm’s information environment determines both how its securities are priced and the extent to which insiders can extract value from private information.

These dynamics manifest differently across industries. The banking sector, for instance, operates under heavy regulatory oversight and strong governance constraints, which reduce the frequency of opportunistic trading. Belkhir (2005) and Lee (2002) demonstrate that insider ownership in banks influences risk-taking behavior, while Del Brio et al. (2018) find that trading intensity increases when governance weakens, suggesting that effective oversight tends to make insider activity more information-driven. In this highly monitored environment, insider transactions are often credible indicators of a firm’s future performance.

Technology firms, by contrast, face high information asymmetry due to their reliance on intangible assets and research and development (R&D) activities. Aboody and Lev (2000) highlight that insiders in these firms benefit from private insights into innovation outcomes, while Xu et al. (2007) and Ciftci and Zhou (2016) show that disclosure of intangible information—such as patents—plays a central role in valuation. Hong et al. (2000) note that information, particularly bad news, spreads more slowly in tech industries, making insider trades harder to interpret, as they may also reflect liquidity or compensation motives rather than a pure information advantage.

In contrast, utility companies operate in a far more transparent setting. As regulated and publicly accountable entities, they maintain stable cash flows and high disclosure standards. Bushman et al. (2004) document that utilities exhibit strong governance and disclosure practices. In contrast, Dyck and Zingales (2004) and Dhaliwal et al. (2014) find that government oversight and corporate social responsibility reporting lower their cost of equity. Consequently, insiders in the utility sector have fewer opportunities to profit from undisclosed information, rendering their trades less informative.

In sum, prior studies show that information asymmetry and disclosure environments influence the profitability of insider trading. However, most of this evidence is based on firm-level characteristics or single-industry settings, and there is limited direct comparison across industries within a unified empirical framework. As a result, it remains unclear whether insider trading signals differ systematically across industries operating under distinct regulatory and informational conditions. This gap is particularly important because industries such as banking, technology, and utilities exhibit substantial differences in transparency, monitoring, and information production. If these differences shape how private information is generated and revealed, then the predictive content of insider trading should vary accordingly across sectors.

Based on this reasoning, we test the following hypothesis:

Hypothesis 1.

Insider purchases in banking firms convey stronger predictive signals of abnormal returns than those in technology and utilities.

2.4. Regulation, Disclosure, and Asymmetric Predictive Power

The extent to which insiders profit from private information depends not only on firm and industry characteristics but also on the surrounding legal and disclosure frameworks. Across markets, insider trading laws have evolved to strike a balance between efficiency, fairness, and investor protection. These institutional differences affect the frequency of insider trading and the amount of information disclosed in their transactions to the public.

From a global standpoint, research highlights that the effectiveness of enforcement, rather than the mere existence of laws, determines regulatory success. Bhattacharya and Daouk (2002) demonstrate that the cost of equity declines only after a country’s first prosecution for insider trading, suggesting that credible enforcement discourages opportunistic behavior and fosters investor trust. Fernandes and Ferreira (2009) reach a similar conclusion: when insider trading laws are actively enforced, price informativeness improves—particularly in developed markets with strong legal institutions. Extending this discussion, Minenna (2003) compares European enforcement models and argues that localized, proactive supervision is essential for detecting and deterring informed trading effectively.

In the United States, the focus of regulation has gradually shifted toward differentiating legitimate trades from opportunistic ones. The introduction of SEC Rule 10b5-1 in 2000 aimed to protect insiders who had prearranged their trades while not in possession of material non-public information. Yet empirical findings suggest that many insiders used these trading plans strategically. Jagolinzer (2009) documents that trades under such plans often follow periods of strong corporate performance and occur before unfavorable news, pointing to deliberate timing. Similarly, Veliotis (2010) contends that some insiders disguise informed trading by misrepresenting their plan intentions. To address these concerns, the 2022 amendment to Rule 10b5-1—analyzed by Kim et al. (2025)—introduced stricter cooling-off periods and banned overlapping or single-trade plans. Although these changes reduced opportunistic activity, they also appeared to dampen price efficiency, illustrating the trade-off between regulatory control and the timely flow of information.

Beyond formal regulation, corporate governance mechanisms further constrain insider behavior. Bettis et al. (2000) report that most U.S. firms implement blackout periods around major announcements, narrowing insiders’ trading windows and tightening bid-ask spreads. These internal restrictions complement legal frameworks by lowering the likelihood of information-driven trades. At the same time, disclosure policies act as another layer of control over how information enters the market. Fried (1997) suggests that mandatory pre-trading disclosure could directly limit insiders’ abnormal profits. In contrast, Cheng and Lo (2006) show that managers often synchronize voluntary forecasts with their own trading, issuing optimistic predictions before buying shares to legitimize such actions and mitigate litigation risk.

Together, these findings reveal that both regulatory enforcement and disclosure quality shape the way insiders trade and how investors interpret their actions. However, the literature also shows that not all insider transactions carry equal informational value. A recurring observation is that purchases are typically far more informative than sales. Lakonishok and Lee (2001) find that insider buying, especially among smaller firms—is followed by positive abnormal returns, while insider selling lacks consistent predictive power. Seyhun (1986) also concludes that most insider profits stem from purchases, and Rozeff and Zaman (1988) demonstrate that, once transaction costs are taken into account, the profitability of mimicking insider sales largely disappears.

This buy–sell asymmetry stems from both behavioral and institutional factors. Buying is generally voluntary and reflects managerial confidence in a firm’s future, whereas sales are often influenced by liquidity needs, tax planning, or diversification motives. Moreover, selling ahead of negative information exposes insiders to higher reputational and regulatory risk, discouraging aggressive behavior. Disclosure practices reinforce this difference: managers tend to release favorable forecasts before buying but rarely do so before selling (Cheng & Lo, 2006). Consequently, insider purchases provide cleaner and more credible signals about firm fundamentals, while insider sales tend to be noisier and more complex to interpret.

In summary, the interaction between regulation, disclosure, and behavioral constraints shapes the informational content of insider trading. Stronger legal oversight and greater transparency can curb opportunism, but they also make insider purchases appear more reliable indicators of genuine private information. These regulatory shifts also suggest that the informativeness of insider trading is not constant over time. When enforcement tightens or disclosure practices improve, insider purchases may stand out as relatively cleaner signals of private information than insider sales.

While prior studies consistently document that insider purchases are more informative than sales, most of this evidence treats the asymmetry as a general market-wide phenomenon. There is limited evidence on whether this asymmetry persists uniformly across institutional settings or depends on variations in regulatory and disclosure environments. In particular, if regulatory constraints and disclosure practices affect insiders’ incentives to trade on private information, the relative informativeness of purchases and sales may vary with the broader institutional context. This motivates a direct test of the asymmetry between insider buying and selling in our empirical setting.

Hypothesis 2.

Insider purchases and insider sales exhibit asymmetric predictive effects on subsequent returns.

2.5. Industry Heterogeneity in Buy–Sell Asymmetry

The degree of buy–sell asymmetry is unlikely to be constant across industries. Firms operate under different levels of regulation, transparency, and information asymmetry, all of which influence how insiders trade and how the market interprets their actions. Industry context, therefore, plays a crucial role in shaping whether insider purchases and sales carry meaningful predictive signals.

In banking and utility sectors, stricter regulatory oversight and higher disclosure standards reduce the scope for opportunistic or liquidity-driven insider trades. Del Brio et al. (2018) find that insider activity in banks becomes more information-driven when governance and monitoring are effective. In contrast, Bushman et al. (2004) document that utility firms exhibit greater transparency due to public accountability and regulatory scrutiny. In such settings, insiders who decide to trade are more likely to do so based on genuine private information, particularly when buying shares. As a result, insider purchases in these industries tend to be cleaner and more reliable indicators of future firm performance.

In contrast, technology firms operate in environments characterized by high information asymmetry and weaker disclosure surrounding intangible assets. Aboody and Lev (2000) show that R&D activities create substantial informational gaps, while Hong et al. (2000) observe that news—especially negative information—spreads more slowly in technology-driven markets. Because insiders in such firms often receive compensation in the form of stock options and face greater uncertainty about valuation, both purchases and sales may reflect mixed motives, such as liquidity needs or portfolio rebalancing, rather than pure informational signals. This complexity reduces the clarity and magnitude of the buy–sell asymmetry observed in the technology sector.

Although prior research establishes that insider purchases are generally more informative than sales, little is known about whether this asymmetry itself varies across industries. Existing studies typically examine buy–sell differences in aggregate settings and do not consider how industry-specific characteristics shape this relationship. This gap is important because industries differ significantly in terms of regulatory oversight, disclosure practices, and information asymmetry. These factors may influence not only the overall informativeness of insider trading but also the relative meaning of purchases and sales.

Therefore, we test whether the buy–sell asymmetry varies systematically across industries:

Hypothesis 3.

The buy–sell asymmetry varies across industries and is stronger in Banks and Utilities than in Technology.

3. Data and Methodology

3.1. Data

The dataset in this paper combines multiple sources. Information on insider trading is obtained from the Thomson Financial Insider Trading database, which reports detailed records of executives’ open-market purchases and sales. This database is accessed through WRDS and provides transaction-level information from Form 4 filings. Data on stock returns and firm-level characteristics come from CRSP and Compustat, including market-adjusted returns, firm size, book-to-market ratio, idiosyncratic volatility, and insider trading intensity (ITI). Bloomberg is used to classify firms into technology, utilities, and banking sectors and to verify industry classifications.

The sample covers U.S.-listed firms from 2005 to 2025. Insider trading data are taken from Form 4 filings, which report the trade date, direction, volume, price, and the insider’s role (CEO, CFO, director). Firms are grouped into technology, utilities, and banking using Bloomberg classifications. The paper excludes option exercises, gifts, and small trades to focus on open-market purchases and sales.

In the baseline analysis, insider transactions are assigned to firm-month observations based on the trade date rather than the filing date. Specifically, all open-market purchases and sales executed within calendar month t are aggregated to construct the insider-trading measures for firm i in month t. This timing convention is chosen because the paper focuses on whether insiders’ trading activity predicts subsequent stock returns, rather than on the short-run market reaction to the public disclosure of Form 4 filings.

The dependent variables are cumulative forward returns over 1-, 3-, and 6-month horizons (Ret(1m), Ret(3m), Ret(6m)). Risk-adjusted abnormal returns (alphas) are also used as a robustness check. The key independent variables capture insider trading activity, including a buy/sell dummy and the net purchase ratio (NPR). Based on Bloomberg industry codes, firms are classified into technology, utilities, and banking sectors. Control variables include firm size, book-to-market ratio, idiosyncratic volatility, insider trading intensity (ITI), and momentum.

Then the paper removes observations with missing values and winsorizes continuous variables at the 1st and 99th percentiles to mitigate the influence of outliers. The final sample consists of approximately 483 firms, 31,399 insider transactions, and 28,461 firm-month observations across the technology, utilities, and banking sectors during 2005~2025.

The construction of the final regression sample proceeds in several steps. First, transaction-level Form 4 data are filtered to retain only open-market purchases and sales and are then aggregated to the firm-month level based on the trade date. Second, these firm-month insider trading measures are merged with a monthly panel constructed from CRSP and Compustat, which provides stock returns and firm characteristics. The merge is performed at the firm-month level using standard identifiers to ensure consistency across datasets. The sample is constructed using standard CRSP and Compustat coverage without imposing survivorship screens, thereby mitigating concerns related to survivorship bias. Forward-looking return variables are then computed over 1-, 3-, and 6-month horizons, and additional firm-level control variables are incorporated.

The aggregation is based on the trade date rather than the filing date, ensuring that the timing of insider trading reflects when the information is incorporated into the market. Firm-month observations are constructed using calendar months, and forward returns are measured starting from the end of each month.

The reduction in the number of usable observations does not primarily arise from the merging procedure itself, but rather from data availability constraints. In particular, missing values in forward returns due to horizon construction, as well as incomplete coverage of firm characteristics such as book-to-market ratio and idiosyncratic volatility, lead to a smaller regression sample. We do not observe systematic evidence that sample attrition is concentrated in any specific industry, suggesting that the reduction in observations is not driven by sector-specific data availability. Importantly, the exclusion of observations due to missing variables does not materially alter the industry composition of the sample. To ensure consistency across specifications, observations with missing key variables are excluded in the baseline analysis. A detailed account of sample attrition at each stage of the data construction process is reported in Appendix A.

To avoid restricting the analysis to periods with observed insider trading activity, the final panel includes firm-month observations with no insider transactions, for which insider trading variables are set to zero. As a result, the increase in observations in the final panel reflects the inclusion of zero-trading firm-months rather than an expansion in underlying transaction data. This approach ensures that the results are not driven by the selective sampling of active trading periods.

Industry classification is based on index membership, with firms drawn from representative sector indices in technology, utilities, and banking. Specifically, the sample is constructed from constituents of widely followed market indices, including the Nasdaq-100 (NDX) for technology firms, the Dow Jones Utility Average (UTY) for utilities, and the KBW Bank Index (BKX) for banking institutions.

This approach defines industry affiliation ex ante through index inclusion rather than relying on firm-level classification codes such as SIC or GICS. Using index-based classification ensures that firms are grouped according to economically meaningful and market-recognized sector definitions commonly adopted in empirical finance research. It also avoids potential inconsistencies arising from changes in firm-level industry coding over time.

Finally, continuous variables are winsorized at the 1st and 99th percentiles to limit the influence of extreme observations. This approach is commonly adopted in the empirical finance and accounting literature to mitigate the impact of outliers while preserving the overall distribution of the data (e.g., Kothari et al., 2005; Frankel & Li, 2004). In the context of this study, the use of the 1% cutoff helps reduce the effect of unusually large insider trading measures and return realizations, which could otherwise disproportionately affect the regression estimates. Importantly, the main results are not sensitive to the treatment of extreme observation.

3.2. Methodology

3.2.1. Research Design

This paper examines whether insider trading activity can predict future stock performance, paying particular attention to industry-level variation and the distinction between buying and selling (Pettit & Venkatesh, 1995; Jenter, 2005; Huddart & Ke, 2007). The analysis focuses on three sectors that differ markedly in their information environments and levels of regulatory scrutiny. For banks, strict capital rules and disclosure requirements do not necessarily weaken the value of insider trades; instead, such conditions may ensure that the trades that do occur are more likely to contain genuine private information (Del Brio et al., 2018). Technology firms, conversely, are marked by intensive R&D, high growth potential, and considerable uncertainty. However, the constant flow of analyst coverage and news often causes markets to absorb signals quickly, thereby reducing the incremental effect of insider purchases (Aboody & Lev, 2000). Utilities operate in tightly regulated markets with predictable earnings streams, giving insiders only limited scope to exploit private information. These contrasts provide a natural setting to examine how insider trading informativeness varies across industries.

The empirical design adopts a monthly panel framework. Insider trading activity observed in month t is used to forecast stock returns over horizons t + 1 to t + h (h ∈ {1,3,6}), corresponding to Ret(1m), Ret(3m), and Ret(6m). The primary explanatory variables are the directional components of the net purchase ratio, defined as NPR+ (= max(NPR, 0)) and NPR− (= min(NPR, 0)), which are further interacted with industry dummies (Dtech, Dutil), with banks as the benchmark group. Control variables include firm size, book-to-market ratio, idiosyncratic volatility, insider trading intensity, Amihud illiquidity, and momentum (Fama & French, 1993; Ang et al., 2006; Amihud, 2002; Jegadeesh & Titman, 1993). Monthly fixed effects absorb common shocks, and standard errors are clustered at the firm level. While these industry classifications capture broad institutional differences, they do not directly measure firm-level transparency, disclosure quality, or information asymmetry. Therefore, the empirical analysis should be interpreted as comparing reduced-form differences across sectors rather than isolating specific information channels.

3.2.2. Variable Construction

The dependent variables are forward-looking stock returns measured over 1-, 3-, and 6-month horizons following insider trading. For firm i in month t, the cumulative return is given by:

$$RE{T}_{i,t\to t+h}={\prod }_{\tau =t+1}^{t+h}\left(1+r_{i,\tau }\right)-1$$

(1)

where $r_{i,\mathsf{\tau }}$ denotes the monthly return of firm i. In the empirical dataset, these measures correspond to Ret(1m), Ret(3m), and Ret(6m).

The key explanatory variable is the net purchase ratio (NPR), constructed at the firm-month level from insiders’ open-market transactions reported in Form 4 filings. For each firm i in month t, we aggregate all shares purchased and all shares sold by insiders across qualifying transactions. The monthly aggregation is based on the month in which the trade is executed, not the month in which the Form 4 is filed. Accordingly, the insider-trading variables reflect insiders’ trading decisions in month t, and the dependent variables are measured over horizons beginning after month t. The NPR is defined as:

$$NP{R}_{i,t}={\displaystyle \frac{Bu{y}_{i,t}-Sel{l}_{i,t}}{Bu{y}_{i,t}+Sel{l}_{i,t}}}$$

(2)

This measure captures the relative imbalance between insider buying and selling within a firm-month. NPR ranges from −1 to 1, where values close to 1 indicate pure buying activity, values close to −1 indicate pure selling activity, and values near zero reflect a balance between purchases and sales.

To examine buy–sell asymmetry, following Huddart and Ke (2007) and Cohen et al. (2012), NPR is decomposed into directional components, where $NP{R}_{i,t}^{+}=\max \left(NP{R}_{i,t},0\right)$ captures net buying intensity and $NP{R}_{i,t}^{-}=\min \left(NP{R}_{i,t},0\right)$ captures net selling intensity. While this decomposition captures the directional intensity of insider trading, it does not distinguish between routine and opportunistic transactions, which may differ in their informational content. As a result, the estimated coefficients reflect an average effect across heterogeneous trading motives.

$$NP{R}_{i,t}^{+}=\max \left(NP{R}_{i,t},0\right),\hspace{1em}NP{R}_{i,t}^{-}=\min \left(NP{R}_{i,t},0\right)$$

(3)

The insider trading variables further interact with sectoral dummies (Dtech, Dutil), using banks as the benchmark group to examine industry differences. A set of control variables is introduced to reduce the influence of confounding factors. Firm size (ln_mktcap) is proxied by the logarithm of market capitalization, while book-to-market equity (BM) is the ratio of book value to market value of equity (Fama & French, 1993). Idiosyncratic volatility (IVOL) is obtained as the standard deviation of residuals from a 12-month rolling market model regression (Ang et al., 2006). ITI is defined as the log of one plus the number of insider trades in a firm-month. It is intended to capture insider trading intensity within the firm-month rather than market liquidity in the microstructure sense. In this setting, higher ITI reflects more concentrated insider trading activity, which may be associated with differences in information flow, monitoring, or trading opportunities across firms and periods. We also include Amihud illiquidity (ILLIQ) as a market-based proxy for trading frictions and stock illiquidity in order to distinguish insider trading intensity from market liquidity conditions more clearly. Measures of recent stock performance are also included as momentum proxies, following Jegadeesh and Titman (1993). All continuous variables are winsorized at the 1st and 99th percentiles to limit the effect of extreme observations. All control variables are measured at time t or are based on information available prior to month t.

3.2.3. Empirical Model

The empirical analysis is conducted using a firm-month panel. In the baseline specification, we focus on firm-month observations with insider trading activity, as NPR is defined based on realized trades. To address potential concerns about selection, we also construct an alternative specification that includes all firm-month observations in the sample period. For months without insider trading activity, NPR is set to zero. This approach allows us to incorporate both the occurrence and the intensity of insider trading within a unified framework.

The baseline specification takes the form:

$$\begin{gathered} RE{T}_{i,t\to t+h}={\alpha }_h+{\beta }_h^{+}NP{R}_{i,t}^{+}+{\beta }_h^{-}NP{R}_{i,t}^{-}+{\gamma }_h^{tech,+}\left(NP{R}_{i,t}^{+}{D}_i^{tech}\right)+ \\ {\gamma }_h^{util,+}\left(NP{R}_{i,t}^{+}{D}_i^{util}\right)+{\gamma }_h^{tech,-}\left(NP{R}_{i,t}^{-}{D}_i^{tech}\right)+{\gamma }_h^{util,-}\left(NP{R}_{i,t}^{-}{D}_i^{util}\right)+{\theta }_h^{\prime }{X}_{i,t}+{\lambda }_t+{\epsilon }_{i,t,h} \end{gathered}$$

(4)

where $RE{T}_{i,t\to t+h}$ is the forward return of firm i over horizon h ∈ {1,3,6}. The key regressors are the directional components of the net purchase ratio, $NP{R}_{i,t}^{+}$, and $NP{R}_{i,t}^{-}$. Industry dummies $D_i^{tech}$ and $D_i^{util}$ capture technology and utilities firms, respectively, with banks as the omitted benchmark category. $X_{i,t}$ is a vector of control variables including firm size, book-to-market ratio, idiosyncratic volatility, insider trading intensity (ITI), Amihud illiquidity (ILLIQ), and momentum. ${\lambda }_t$ denotes month fixed effects, and ${\epsilon }_{i,t,h}$ is the error term. Standard errors are clustered at the firm level. This specification enables three sets of tests. First, the baseline coefficient ${\beta }_h^{+}$ captures the informativeness of insider purchases in banks (the benchmark group). Comparisons with the interaction terms ${\gamma }_h^{tech,+}$ and ${\gamma }_h^{util,+}$ indicate whether purchases in technology and utility firms are weaker compared to banks. Second, comparing the estimates of ${\beta }_h^{+}$ and ${\beta }_h^{-}$ tests the asymmetry between insider purchases and sales. Finally, the interaction coefficients ${\gamma }_h^{tech,-}$ and ${\gamma }_h^{util,-}$ capture whether the buy–sell asymmetry varies systematically across industries.

Across all specifications, the regressions include month fixed effects to absorb common shocks, while firm characteristics mitigate omitted-variable bias. Additionally, as a robustness check, we estimate the model using the full firm-month panel, setting NPR to zero in non-trading months. The results from the full-sample specification are consistent with the baseline findings, suggesting that the main conclusions are not driven by conditioning on trade months.

Because the 3-month and 6-month forward returns are constructed from overlapping monthly windows, adjacent observations may share components of future returns. In the baseline specification, we address within-firm serial dependence by clustering standard errors at the firm level and include month fixed effects to absorb common time shocks. As an additional robustness check, we re-estimate the main models using non-overlapping return windows for the 3-month and 6-month horizons. The results remain qualitatively consistent with the baseline findings, indicating that the main conclusions are not driven by dependence on overlapping returns. Month fixed effects absorb common shocks across firms in a given period, while firm-clustered standard errors allow for within-firm serial dependence. As a robustness check, the paper also considers specifications with stricter fixed-effects structures to assess whether the baseline results are sensitive to sector-specific time variation.

4. Results and Discussion

4.1. Summary Statistics

Table 1. Descriptive data.

Variable	Obs	Mean	Std.Dev.	Min	P25	Median	P75	Max
Ret (1m)	27,346	0.015	0.097	−0.711	−0.036	0.014	0.062	1.052
Ret (3m)	27,062	0.047	0.174	−0.819	−0.044	0.039	0.126	2.695
Ret (6m)	26,636	0.096	0.27	−0.92	−0.045	0.075	0.203	4.214
NPR	14,211	−0.5348	0.0621	−1	−1	−1	−0.2342	1
Size	28,461	10.178	1.482	4.087	9.251	10.157	11.104	15.283
BM	24,196	0.444	0.452	−0.251	0.121	0.316	0.649	10.533
IVOL (12m)	18,406	0.068	0.039	0.007	0.044	0.058	0.08	0.391
ITI	15,099	1.794	1.111	0.693	1.099	1.609	2.303	8.149
ILLIQ	8428	0.000	0.000	0.000	0.000	0.000	0.000	0.009

Note: This table presents descriptive statistics for the variables used in the analysis. The dependent variables are forward-looking stock returns measured over 1-, 3-, and 6-month horizons. The primary explanatory variable, NPR, reflects the net purchase ratio of insiders. Control variables include firm size (logarithm of market capitalization), book-to-market equity (BM), idiosyncratic volatility (IVOL, estimated from a 12-month market model), insider trading intensity (ITI, defined as the log of one plus the number of insider trades), and Amihud illiquidity (ILLIQ). The dataset comprises up to 28,461 firm-month observations. Detailed variable definitions and data sources are provided in Appendix B.

presents descriptive statistics for the sample. Average forward returns (Ret) amount to 1.5%, 4.7%, and 9.6% over the 1-, 3-, and 6-month horizons, respectively, with volatility increasing over longer periods (standard deviations of 0.097, 0.174, and 0.270). The insider trading indicator, NPR, has a mean of −0.5348 and a distribution that is skewed toward negative values. This reflects that many firm-month observations involve only insider sales, resulting in NPR values close to −1, a common feature of insider trading data where sales are more frequent and often driven by liquidity or diversification motives. Firm characteristics vary substantially across the sample: the average log market capitalization (Size) is 10.178, the mean book-to-market ratio (BM) is 0.444, idiosyncratic volatility (IVOL (12m)) averages 0.068 with a dispersion of 0.039, and the insider-trading intensity measure (ITI) records an average of 1.794. All continuous variables are winsorized at the 1st and 99th percentiles to mitigate the influence of outliers.

Table 2. Pearson correlation coefficients between regression variables.

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
(1) Ret (1m)	1
(2) Ret (3m)	0.58 ***	1
(3) Ret (6m)	0.44 ***	0.71 ***	1
(4) NPR	−0.03 ***	−0.04 ***	−0.05 ***	1
(5) Size	−0.06 ***	−0.08 ***	−0.10 ***	0.17 ***	1
(6) BM	0.02 *	−0.01	−0.01	0.08 ***	−0.17 ***	1
(7) IVOL (12m)	0.12 ***	0.14 ***	0.20 ***	−0.15 ***	−0.14 ***	−0.08 ***	1
(8) ITI	0.03 ***	0.03 ***	0.03 **	−0.26 ***	0.16 ***	−0.21 ***	0.22 ***	1
(9) ILLIQ	0.06 ***	0.07 ***	0.09 ***	−0.08 ***	−0.38 ***	0.08 ***	0.17 ***	−0.03 **	1

Note: This table presents Pearson correlation coefficients among the variables used in the empirical analysis. The dependent variables are forward returns over 1-, 3-, and 6-month horizons, while NPR denotes the insider net purchase ratio. Control variables include firm size (log market capitalization), book-to-market equity (BM), idiosyncratic volatility (IVOL, based on a 12-month market model), and insider trading intensity (ITI). The sample comprises up to 28,461 firm-month observations. Detailed variable definitions and data sources are provided in Appendix B. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

illustrates the Pearson correlations among key variables. Short-, medium-, and long-term (1-, 3-, and 6-month) returns are positively associated (correlations of 0.58 between 1- and 3-month, 0.44 between 1- and 6-month, and 0.71 between 3- and 6-month returns). The insider signal (NPR) is weakly related to near-term returns (ρ = 0.03 *) but negatively associated with medium- and long-term returns (ρ = −0.05 *). This unconditional correlation reflects the dominance of insider sales in the sample and does not account for the directional decomposition of NPR used in the regression analysis. Control variables show modest correlations—such as Size–BM (−0.17 ***); Size–IVOL (−0.14 ***); BM–ITI (−0.21 ***); and IVOL–ITI (0.22 ***)—indicating that multicollinearity is unlikely to bias the regression estimates. Because NPR is bounded between −1 and 1 and exhibits a skewed distribution, its economic interpretation is best understood in terms of directional movements and relative changes rather than raw coefficient magnitudes.

4.2. Baseline Regression Results

Table 3. Insider trading signals and future returns.

Variable	Ret(1m)	Ret(3m)	Ret(6m)
NPR+ (Buy)	−0.001	0.004 ***	0.006 ***
	(−1.93)	(7.36)	(4.79)
NPR+ × Tech	0.001 *	−0.004 ***	−0.006 ***
	(2.01)	(−7.43)	(−4.75)
NPR+ × Utilities	−0.006 *	−0.026 ***	0.019 **
	(−2.38)	(−6.25)	(3.17)
NPR− (Sell)	0.000 *	0.000 ***	0.001 ***
	(1.84)	(7.42)	(6.21)
NPR− × Tech	−0.000	−0.001 **	−0.001 **
	(−0.61)	(−3.10)	(−2.67)
NPR− × Utilities	0.000	−0.000 ***	−0.001 ***
	(1.86)	(−4.72)	(−3.95)
Size	−0.001	−0.004 *	−0.008
	(−1.75)	(−2.10)	(−1.88)
BM	−0.001	−0.0130 ***	−0.026 **
	(−0.45)	(−3.35)	(−2.85)
IVOL (12m)	0.234 ***	0.586 ***	1.29 ***
	(4.95)	(6.83)	(7.75)
ILLIQ	6.8284	9.661	21.967 *
	(1.29)	(1.59)	(2.20)
Month FE	Yes	Yes	Yes
Observations	8508	8439	8331
Cluster SE	Firm	Firm	Firm
Adj. R2	0.277	0.231	0.206

Note: This table presents baseline OLS regression results linking insider trading activity to subsequent stock returns over 1-, 3-, and 6-month horizons. The key explanatory variables are the positive and negative components of NPR, along with their interactions with Technology and Utilities, with Banking as the omitted benchmark sector. All regressions include month fixed effects and firm-clustered standard errors. t-statistics are shown in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

reports the results of panel regressions examining the predictive relation between insider trading activity and future stock returns across different time horizons. Each model includes firm-level controls such as Size, BM, IVOL (12m), and Amihud illiquidity (ILLIQ), together with the insider-trading variables defined earlier.

The results show that insider purchases are positively associated with future stock returns, especially at the 3-month and 6-month horizons. The coefficient on insider purchases (NPR⁺) is positive and statistically significant for the 3-month and 6-month horizons (0.004 and 0.006, respectively, both at the 1% level), suggesting that insider buying generally predicts higher future returns.

The interaction terms indicate substantial cross-industry differences. The coefficient on NPR⁺ × Tech is negative and highly significant, while the interaction for utilities is negative in the short term but turns positive at the 6-month horizon. These results indicate that insider purchases in banking firms have the strongest predictive content, whereas technology firms exhibit the weakest buy-signal effect. The evidence for utilities is more mixed across horizons. Economically, these findings are consistent with cross-industry differences that may reflect variation in firms’ information environments.

In addition to statistical significance, it is important to consider the economic magnitude of these effects. Because the insider-trading variables are bounded between −1 and 1, the magnitude is best interpreted as directional changes in insider-trading intensity rather than as one-unit movements. The estimates suggest that stronger net insider purchases are associated with economically meaningful increases in future returns, particularly in banking firms, while the corresponding effects in technology firms are substantially weaker. For example, movements from stronger net selling toward stronger net buying are associated with economically meaningful increases in 3-month returns for banking firms.

In the banking sector, strict supervision and extensive disclosure obligations make insider trading relatively infrequent, yet such transactions tend to be more informative when they do occur. This pattern is consistent with the possibility that insider trades in banking firms reflect more concentrated or less frequently revealed private information (Del Brio et al., 2018).

Technology firms operate in environments with intensive R&D, rapid innovation, and frequent information releases. Because information circulates quickly in these markets, any private signals that insiders possess are more likely to be absorbed by investors, reducing the persistence of return predictability (Aboody & Lev, 2000).

In the case of utilities, regulatory oversight and pricing constraints tend to moderate the speed at which new information is incorporated into share prices. Consequently, insider transactions in this sector may exert a slower and more gradual influence on subsequent market performance than those in less regulated industries.

Turning to insider sales, the coefficient on NPR⁻ is statistically significant in the medium- and long-term horizons, but the economic magnitude is smaller than that of purchases. This suggests that insider selling conveys weaker and less consistent signals about future performance. From an economic perspective, shifts toward net buying are associated with larger changes in future returns than comparable shifts toward net selling, reinforcing the view that insider purchases provide a stronger and more economically meaningful signal than sales.

For insider sales, the interpretation is more nuanced. While sales are often associated with liquidity needs, tax considerations, or portfolio rebalancing, they may also reflect insiders’ private information about future firm performance. In the current specification, the sell measure aggregates all insider sales at the firm-month level, thereby combining both routine and opportunistic transactions. As a result, the estimated coefficient on NPR⁻ reflects an average across heterogeneous trading motives. The relatively weaker predictive power of sales should therefore not be interpreted as evidence that insider selling is uninformative, but rather that its informational content is more difficult to isolate in aggregated data.

The interaction terms for NPR⁻ further suggest that buy–sell asymmetry varies across industries. The coefficients on NPR⁻ × Tech are negative and significant, indicating that insider selling in technology firms is less predictive. The interaction terms for utilities are also significant at the medium- and long-term horizons, suggesting that the interpretation of insider sales differs across sectors. Overall, these results indicate that the magnitude of buy–sell asymmetry is not uniform across industries and appears strongest in banking and utilities but weaker in technology firms.

Overall, the results suggest that insider buying is generally more informative than selling, and that this informational advantage is particularly strong in more regulated and transparent industries such as banking. In contrast, industries characterized by higher uncertainty and faster information diffusion, such as technology, exhibit a weaker linkage between insider trading and future returns. The next section examines whether these patterns differ across regulatory regimes following the 2022 amendments to Rule 10b5-1.

4.3. Additional Results: Regulatory Reform and the Informativeness of Insider Trades

The paper further estimates the main specification using risk-adjusted alphas (FF3) as the dependent variable and splits the sample around the SEC’s Rule 10b5-1 reform. The goal is to examine whether the signal content of insider trading differs following changes in trading plans and disclosure rules (Huddart & Ke, 2007; Del Brio et al., 2018). Because the current data do not distinguish Rule 10b5-1 plan trades from other insider transactions and do not provide a clean treatment-control structure, the analysis should be interpreted as descriptive evidence on differences across periods rather than a causal test of the amended rule.

The specification is written as:

$${\alpha }_{i,t\to t+h}={\beta }_0+{{\beta }_1\mathit{NPR}}_{i,t}^{+}+{{\beta }_2\mathit{NPR}}_{i,t}^{-}+{\beta }_3\left({\mathit{NPR}}_{i,t}^{+}\times \mathit{Post}\mathsf{2023}\right)+{\beta }_4\left({\mathit{NPR}}_{i,t}^{-}\times \mathit{Post}\mathsf{2023}\right)+{\theta }^{\prime }{X}_{i,t}+{\lambda }_t+{\epsilon }_{i,t,h}$$

(5)

where ${\mathsf{\alpha }}_{i,t\to t+h}$ denotes forward abnormal returns from the Fama–French three-factor model, and Post2023 is an indicator equal to one for observations after 2023 and zero otherwise.

Pre-reform (2005–2022): Panel A shows limited evidence that insider sales are associated with lower future alphas at the three-month horizon, while purchase signals are weak. Interaction terms suggest that utilities exhibit the strongest pattern: NPR−×Utilities is positive and significant across horizons, consistent with sales in stable, regulated firms and indicating that markets gradually incorporate this information.

Post-reform (2023–2025): Panel B shows weaker statistical associations in the post-2023 period. Coefficients on purchase measures remain insignificant, and industry interactions lose power. Panel C confirms this pattern through the interaction design—NPR⁻ × Post2023 is negative and significant at 1-, 3-, and 6-month horizons (approximately −0.005, −0.017, −0.033 after scaling), while ${\mathit{NPR}}_{i,t}^{+} \times$ Post2023 stays insignificant. Economically, this pattern is consistent with weaker sell-side associations in the post-2023 period, whereas buy-side activity appears largely unchanged.

In sum,

Table 4. Subsample: Before and After the Rule 10b5-1 Reform.

Panel A. Pre-Reform
Variables	(1) α (FF3, 1m)	(2) α (FF3, 3m)	(3) α (FF3, 6m)
NPR+	−0.001	0.000	−0.001
	(−1.35)	(0.56)	(−0.44)
NPR+ × Tech	0.001	−0.000	0.001
	(1.43)	(−0.44)	(0.46)
NPR+ × Utilities	−0.010 **	−0.027	−0.044
	(−2.65)	(−1.89)	(−1.86)
NPR−	0.000	−0.001 ***	−0.012
	(0.83)	(−4.01)	(−0.96)
NPR− × Tech	0.000	0.001 *	0.014
	(0.75)	(2.33)	(1.08)
NPR− × Utilities	0.007 ***	0.024 ***	0.052 ***
	(10.59)	(10.78)	(3.76)
Size	−0.001	0.002	0.007
	(−1.12)	(0.72)	(1.05)
BM	0.008	0.025	0.060 *
	(1.47)	(1.84)	(2.07)
IVOL (12m)	0.019	0.543	1.031
	(0.10)	(0.92)	(0.96)
RET (1m)	−0.045	0.921 ***	0.079
	(−1.37)	(9.78)	(0.50)
RET (3m)	0.052 **	0.072	0.597 ***
	(2.71)	(1.51)	(7.14)
RET (6m)	−0.025 *	−0.063 *	0.225 ***
	(−2.45)	(−2.41)	(4.26)
RET (12m)	0.014	0.016	−0.067
	(1.27)	(0.45)	(−1.10)
MOM_{12−2}	−0.025 **	−0.042	−0.025
	(−3.11)	(−1.59)	(−0.61)
VOL (12m)	−0.083	−0.771	−1.655
	(−0.62)	(−1.80)	(−1.90)
ILLIQ	−7.311	−17.098	−57.116
	(−0.48)	(−0.39)	(−0.63)
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	6914	6914	6869
Adj. R2	0.990	0.992	0.993
Panel B. Post-Reform
Variables	(1) α (FF3, 1m)	(2) α (FF3, 3m)	(3) α (FF3, 6m)
NPR+	−0.468	−1.812	−2.698
	(−1.64)	(−1.71)	(−1.91)
NPR+ × Tech	0.510	1.839	2.652
	(1.76)	(1.71)	(1.85)
NPR+ × Utilities	−0.841	−1.745	−3.217
	(−1.62)	(−1.05)	(−1.16)
NPR−	0.000	0.004	−0.006
	(0.05)	(0.19)	(−0.21)
NPR− × Tech	−0.003	−0.013	−0.014
	(−0.41)	(−0.66)	(−0.46)
NPR− × Utilities	0.116	0.188	0.697
	(0.79)	(0.55)	(1.10)
Size	−0.008	−0.014	−0.046
	(−1.14)	(−0.72)	(−1.30)
BM	0.013	0.031	0.027
	(0.25)	(0.18)	(0.09)
IVOL (12m)	0.647	1.510	2.311
	(1.03)	(0.78)	(0.68)
RET (1m)	−0.079	0.883 ***	0.246
	(−0.96)	(3.97)	(0.67)
RET (3m)	0.027	0.068	0.487 *
	(0.50)	(0.49)	(2.39)
RET (6m)	−0.039	−0.125	0.254
	(−1.03)	(−1.08)	(1.25)
RET (12m)	−0.014	−0.068	−0.182
	(−0.41)	(−1.04)	(−1.60)
MOM_{12−2}	0.012	0.028	0.046
	(0.30)	(0.35)	(0.38)
VOL (12m)	0.070	0.244	0.008
	(0.12)	(0.13)	(0.00)
ILLIQ	85.776	293.718	−18.905
	(0.30)	(0.33)	(−0.01)
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	1514	1514	1505
Adj. R2	0.984	0.984	0.985
Panel C. Pooled × Post Reform
Variables	(1) α (FF3, 1m)	(2) α (FF3, 3m)	(3) α (FF3, 6m)
NPR+	−0.000	0.000	−0.000
	(−0.07)	(0.03)	(−0.40)
NPR−	0.002	0.005	0.010
	(1.19)	(1.18)	(1.20)
NPR+ × Post2023	0.021	−0.056	−0.159
	(1.07)	(−0.83)	(−1.25)
NPR− × Post2023	−0.005 **	−0.017 **	−0.033 **
	(−2.89)	(−2.99)	(−3.06)
Size	−0.002	−0.000	−0.001
	(−1.56)	(−0.07)	(−0.08)
BM	0.007	0.021	0.047
	(0.90)	(0.85)	(1.12)
IVOL (12m)	0.196	0.867	1.444
	(0.99)	(1.40)	(1.36)
RET (1m)	−0.045	0.931 ***	0.148
	(−1.35)	(10.84)	(0.98)
RET (3m)	0.042 *	0.058	0.547 ***
	(2.30)	(1.24)	(6.92)
RET (6m)	−0.019	−0.055	0.259 ***
	(−1.56)	(−1.87)	(4.22)
RET (12m)	0.003	−0.008	−0.097
	(0.23)	(−0.21)	(−1.53)
MOM_{12−2}	−0.017	−0.031	−0.016
	(−1.37)	(−1.02)	(−0.35)
VOL (12m)	−0.051	−0.549	−1.221
	(−0.35)	(−1.15)	(−1.43)
ILLIQ	−10.762	−25.411	−73.894
	(−0.68)	(−0.55)	(−0.78)
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	8428	8428	8374
Adj. R2	0.991	0.992	0.994

Note: This table presents OLS regression results testing whether the predictive power of insider trading changed following the SEC’s Rule 10b5-1 reform. Panel A reports estimates for the Pre-Reform period (2005–2022), Panel B for the Post-Reform period (2023–2025), and Panel C pools both periods and includes interactions between NPR variables and a Post2023 dummy. The dependent variables are risk-adjusted abnormal returns (α) from the Fama–French three-factor model, measured over 1-, 3-, and 6-month horizons. All regressions include month fixed effects, with standard errors clustered at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

shows that before the reform, utilities’ sell-signals were the most informative, but after the reform, sell-side predictability weakened materially. These findings are consistent with the view that enhanced oversight may be associated with changes in sell-side trading behavior (Huddart & Ke, 2007; Del Brio et al., 2018). Future research that distinguishes between plan-based and non-plan trades may provide clearer evidence on the mechanisms underlying these patterns.

4.4. Additional Results: Signal Intensity and Alternative NPR Definitions

To examine whether the findings depend on how insider trading activity is measured,

Table 5. Alternative Variable Definitions—Different NPR Thresholds.

Panel A. Buy/Sell Dummies (NPR > 0 or <0)
Variables	(1) α (FF3, 1m)	(2) α (FF3, 3m)	(3) α (FF3, 6m)
Buy Dummy	−0.004	−0.0180	−0.038
	(−0.69)	(−0.89)	(−1.25)
Sell Dummy	−0.009	−0.0200	−0.041
	(−1.85)	(−1.34)	(−1.51)
Size	−0.002	0.000	−0.002
	(−1.31)	(0.02)	(−0.23)
BM	0.011	0.021	0.045
	(1.19)	(0.70)	(0.83)
IVOL (12m)	0.126	0.731	1.127
	(0.87)	(1.58)	(1.38)
RET (1m)	−0.067 **	0.822 ***	−0.023
	(−2.74)	(18.63)	(−0.30)
RET (3m)	0.019	0.043	0.658 ***
	(1.53)	(1.24)	(11.52)
RET (6m)	−0.027 **	−0.064 *	0.194 ***
	(−3.07)	(−2.50)	(4.08)
RET (12m)	0.011	−0.009	−0.103 **
	(1.14)	(−0.33)	(−2.71)
MOM_{12−2}	−0.014	−0.013	0.026
	(−1.93)	(−0.77)	(0.98)
VOL (12m)	0.142	0.018	−0.409
	(0.86)	(0.03)	(−0.44)
ILLIQ	−5.506	−8.761	−29.004
	(−1.12)	(−0.56)	(−0.78)
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	16,155	16,155	16,040
Adj. R2	0.989	0.990	0.992
Panel B. Top/Bottom Terciles (Top 33%/Bottom 33%)
Variables	(1) α (FF3, 1m)	(2) α (FF3, 3m)	(3) α (FF3, 6m)
High Buy (Top 33%)	−0.003	−0.018	−0.033
	(−0.89)	(−1.71)	(−1.96)
High Sell (Bottom 33%)	−0.010	−0.017	−0.038
	(−1.27)	(−0.72)	(−0.90)
Size	−0.003	−0.000	−0.003
	(−1.57)	(−0.05)	(−0.34)
BM	0.011	0.022	0.047
	(1.21)	(0.74)	(0.86)
IVOL (12m)	0.130	0.739	1.144
	(0.91)	(1.62)	(1.42)
RET (1m)	−0.067 **	0.820 ***	−0.026
	(−2.73)	(18.21)	(−0.34)
RET (3m)	0.019	0.043	0.658 ***
	(1.51)	(1.21)	(11.49)
RET (6m)	−0.026 **	−0.064 *	0.194 ***
	(−3.06)	(−2.52)	(4.06)
RET (12m)	0.011	−0.009	−0.103 **
	(1.13)	(−0.34)	(−2.73)
MOM_{12−2}	−0.014	−0.013	0.026
	(−1.91)	(−0.77)	(1.00)
VOL (12m)	0.142	0.018	−0.411
	(0.86)	(0.03)	(−0.45)
ILLIQ	−5.537	−8.571	−28.775
	(−1.13)	(−0.55)	(−0.77)
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	16,155	16,155	16,040
Adj. R2	0.989	0.990	0.992
Panel C. Top/Bottom Quartiles (Top 25%/Bottom 25%)
Variables	(1) α (FF3, 1m)	(2) α (FF3, 3m)	(3) α (FF3, 6m)
High Buy (Top 25%)	−0.003	−0.020 *	−0.041 **
	(−1.08)	(−2.20)	(−2.71)
High Sell (Bottom 25%)	−0.011	−0.017	−0.037
	(−1.16)	(−0.56)	(−0.69)
Size	−0.003	−0.000	−0.003
	(−1.59)	(−0.06)	(−0.33)
BM	0.011	0.023	0.048
	(1.23)	(0.76)	(0.90)
IVOL (12m)	0.131	0.741	1.151
	(0.93)	(1.64)	(1.43)
RET (1m)	−0.068 **	0.819 ***	−0.029
	(−2.74)	(18.14)	(−0.38)
RET (3m)	0.019	0.042	0.656 ***
	(1.50)	(1.20)	(11.50)
RET (6m)	−0.027 **	−0.064 *	0.193 ***
	(−3.07)	(−2.53)	(4.03)
RET (12m)	0.011	−0.008	−0.102 **
	(1.15)	(−0.32)	(−2.67)
MOM_{12−2}	−0.014	−0.014	0.026
	(−1.93)	(−0.80)	(0.96)
VOL (12m)	0.141	0.015	−0.417
	(0.86)	(0.03)	(−0.46)
ILLIQ	−5.531	−8.510	−28.568
	(−1.12)	(−0.55)	(−0.76)
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	16,155	16,155	16,040
Adj. R2	0.989	0.990	0.992

Note: This table presents OLS regression results using alternative definitions of insider trading activity to test the robustness of the main findings. Panel A replaces continuous NPR⁺ and NPR⁻ measures with buy and sell dummy variables, while Panels B and C classify firms into the top/bottom terciles and quartiles of the NPR distribution, respectively. The dependent variables are risk-adjusted abnormal returns (α) from the Fama–French three-factor model, measured over 1-, 3-, and 6-month horizons. All regressions include month fixed effects, with standard errors clustered at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10%.

re-estimates the model using alternative definitions of the insider signal. The baseline specification can be expressed as:

$${\mathsf{\alpha }}_{i,t\to t+h}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{Signal}}_{i,t}+{\mathsf{\theta }}^{\prime }{X}_{i,t}+{\mathsf{\lambda }}_t+{\mathsf{\epsilon }}_{i,t,h}$$

(6)

where ${\mathsf{\alpha }}_{i,t\to t+h}$ denotes the Fama–French three-factor alpha, and ${\mathrm{Signal}}_{i,t}$ is defined using different net purchase ratio transformations (NPR).

Panel A replaces the continuous NPR measure with buy and sell dummies that equal 1 when insiders are net buyers (or sellers) in a given month. The coefficients on both dummies are negative but insignificant, implying that simple directional indicators provide limited evidence of predictive content. Consistent with Cohen et al. (2012), this suggests that the magnitude of insider trading, not merely its sign, conveys meaningful private information.

Panels B and C further group firms into the top/bottom terciles and quartiles of the NPR distribution to capture the strength of insider trading intensity. Results show that extreme buying (High Buy) is followed by modest negative alphas, significant at longer horizons (−0.020 and −0.041 for 3- and 6-month windows). These magnitudes correspond to economically modest effects relative to average returns reported in Table 1.

The pattern presented suggests that extreme insider buying may be associated with short-term return reversals when returns are measured by risk-adjusted alphas, whereas baseline results based on raw returns continue to show positive predictive effects. (Seyhun, 1986; Lakonishok & Lee, 2001). In contrast, extreme selling (High Sell) remains largely insignificant, supporting the notion that insider sales are more often driven by diversification or liquidity motives than by negative private information (Jenter, 2005).

These results confirm that the main inferences are not sensitive to how insider activity is defined. The informational value of insider trading depends on trading intensity, reinforcing the robustness of the baseline findings.

4.5. Robustness Checks: Alternative Dependent Variable—Risk-Adjusted Alpha

To ensure that market-wide risk exposures do not drive the predictive power of insider trading,

Table 6. Alternative Dependent Variable—Risk-Adjusted Alpha.

Panel A. No Interaction Terms
Variables	(1) α (FF3, 1m)	(2) α (FF3, 3m)	(3) α (FF3, 6m)
NPR+	0.000	0.000	0.000
	(−0.11)	(−0.01)	(−0.46)
NPR−	0.001	0.004	0.008
	(1.03)	(0.99)	(0.96)
Size	−0.002	−0.000	−0.000
	(−1.57)	(−0.08)	(−0.08)
BM	0.007	0.022	0.048
	(0.93)	(0.87)	(1.15)
IVOL (12m)	0.240	1.017	1.747
	(1.16)	(1.55)	(1.53)
RET (1m)	−0.039	0.947 ***	0.180
	(−1.18)	(10.85)	(1.19)
RET (3m)	0.047 *	0.078	0.586 ***
	(2.51)	(1.52)	(6.94)
RET (6m)	−0.017	−0.049	0.272 ***
	(−1.36)	(−1.58)	(4.18)
RET (12m)	−0.003	−0.029	−0.140
	(−0.16)	(−0.62)	(−1.76)
MOM_{12−2}	−0.013	−0.017	0.011
	(−1.04)	(−0.55)	(0.22)
VOL (12m)	−0.077	−0.633	−1.392
	(−0.51)	(−1.30)	(−1.60)
ILLIQ	−10.744	−25.250	−73.232
	(−0.68)	(−0.55)	(−0.78)
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	8428	8428	8374
Adj. R2	0.991	0.992	0.994
Panel B. With Industry Interaction Terms (Tech, Utilities)
Variables	(1) α (FF3, 1m)	(2) α (FF3, 3m)	(3) α (FF3, 6m)
NPR+	−0.001	0.000	−0.002
	(−1.21)	(−0.10)	(−0.50)
NPR+ × Tech	0.001	0.000	0.002
	(1.21)	(0.10)	(0.49)
NPR+ × Utilities	−0.011 **	−0.030 *	−0.049 *
	(−3.07)	(−2.10)	(−2.08)
NPR−	0.000	−0.001 **	−0.007
	(0.81)	(−2.72)	(−0.73)
NPR− × Tech	0.000	0.000	0.006
	(−0.43)	(0.06)	(0.61)
NPR− × Utilities	0.007 ***	0.024 ***	0.046 ***
	(10.56)	(10.75)	(4.41)
Size	−0.002	−0.000	−0.000
	(−1.59)	(−0.09)	(−0.06)
BM	0.010	0.028	0.060
	(1.25)	(1.16)	(1.42)
IVOL (12m)	0.252	1.051	1.779
	(1.24)	(1.65)	(1.60)
RET (1m)	−0.038	0.951 ***	0.179
	(−1.15)	(11.11)	(1.20)
RET (3m)	0.045 *	0.072	0.582 ***
	(2.50)	(1.52)	(7.34)
RET (6m)	−0.019	−0.056	0.261 ***
	(−1.59)	(−1.89)	(4.15)
RET (12m)	−0.002	−0.028	−0.139
	(−0.14)	(−0.63)	(−1.86)
MOM_{12−2}	−0.012	−0.01	0.014
	(−0.99)	(−0.48)	(0.31)
VOL (12m)	−0.114	−0.750	−1.567
	(−0.79)	(−1.59)	(−1.84)
ILLIQ	−11.505	−27.78	−77.606
	(−0.73)	(−0.61)	(−0.83)
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	8428	8428	8374
Adj. R2	0.991	0.992	0.994

Note: This table presents OLS regression results using risk-adjusted abnormal returns (α), estimated from the Fama–French three-factor model, as the dependent variable. Panel A reports baseline results without industry interactions, while Panel B incorporates interactions between NPR measures and Tech and Utilities industry dummies. The dependent variables are forward alphas measured over 1-, 3-, and 6-month horizons. All regressions include month fixed effects, with standard errors clustered at the firm level. t-statistics are shown in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

re-estimates the baseline model using risk-adjusted abnormal returns (α) as the dependent variable. These alphas are derived from the Fama–French three-factor model, following the specification:

$$R_{i,t}-R_{f,t}={\mathsf{\alpha }}_i+{\mathsf{\beta }}_{MKT}\left(R_{MKT,t}-R_{f,t}\right)+{\mathsf{\beta }}_{SMB}SM{B}_t+{\mathsf{\beta }}_{HML}HM{L}_t+{\mathsf{\epsilon }}_{i,t}$$

(7)

The estimated ${\mathsf{\alpha }}_{i,t}$ represents each firm’s idiosyncratic excess return after controlling for size, value, and market risk factors. These alphas are then used as the dependent variable in the main regression framework to test whether insider activity predicts abnormal performance beyond systematic risk components:

$$\begin{gathered} {\mathsf{\alpha }}_{i,t\to t+h}={\mathsf{\beta }}_0+{\mathsf{\beta }}_{1}NP{R}_{i,t}^{+}+{\mathsf{\beta }}_{2}NP{R}_{i,t}^{\mp }{\mathsf{\gamma }}_1\left(NP{R}_{i,t}^{+}\times D_i^{tech}\right)+{\mathsf{\gamma }}_2\left(NP{R}_{i,t}^{+}\times D_i^{util}\right) \\ +{\mathsf{\gamma }}_3\left(NP{R}_{i,t}^{-}\times D_i^{tech}\right)+{\mathsf{\gamma }}_4\left(NP{R}_{i,t}^{-}\times D_i^{util}\right)+{\mathsf{\theta }}^{\prime }{X}_{i,t}+{\mathsf{\lambda }}_t+{\mathsf{\epsilon }}_{i,t,h} \end{gathered}$$

(8)

All variable definitions remain consistent with the main model, α denoting Fama–French adjusted excess returns, $X_{i,}$ representing firm-level controls, and ${\lambda }_t$ indicating month fixed effects.

Panel A of Table 6 reports results without industry interaction terms. Both $NP{R}^{+}$ and $NP{R}^{-}$ show small and statistically insignificant coefficients across 1-, 3-, and 6-month horizons, suggesting that raw insider activity alone does not explain excess returns once common risk factors are accounted for.

Panel B incorporates industry interactions with the technology and utilities sectors. The results reveal a clear cross-industry contrast. For utilities, the interaction coefficients ($NP{R}^{+}\times Utility$) are negative and significant across horizons (−0.011, −0.030, −0.049), implying that insider purchases in regulated industries tend to be followed by lower alphas, possibly reflecting short-term overpricing. In contrast, the coefficients for $NP{R}^{-}\times D_{util}$ are positive and highly significant, indicating that insider selling in utilities conveys strong negative information about future risk-adjusted performance. Both purchase and sale interactions remain small and statistically insignificant for technology firms, consistent with earlier evidence that informational advantages are quickly arbitrated away in high-information-flow environments (Aboody & Lev, 2000).

Overall, the findings confirm that the study’s main conclusions are not sensitive to the choice of performance metric. The insider–return relationship persists even after adjusting for Fama–French risk factors, reinforcing the view that insider trades carry incremental information not captured by standard asset-pricing models.

4.6. Robustness Checks: Alternative Fixed-Effects Specifications

To ensure that the baseline findings are not driven by unobserved heterogeneity across firms or industries,

Table 7. Insider Trading and Future Returns: Robustness with Fixed Effects Specifications.

Panel A (Industry × Month FE)
Variable	Ret (1m)	Ret (3m)	Ret (6m)
NPR+	0.000	−0.000	0.000
	(1.25)	(−1.20)	(1.10)
NPR−	−0.000	−0.000	−0.000
	(−0.12)	(−1.28)	(−0.99)
Size	−0.002 **	−0.005 **	−0.009 **
	(−2.09)	(−2.15)	(−2.06)
BM	0.002	−0.004	−0.002
	(0.40)	(−0.55)	(−0.17)
IVOL (12m)	0.205 ***	0.548 ***	1.193 ***
	(3.83)	(5.45)	(6.13)
ILLIQ	7.548 **	10.385 **	22.291 **
	(2.10)	(2.05)	(2.35)
Industry × Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	8508	8439	8331
Adj. R2	0.3563	0.3163	0.2778
Panel B. Firm Fixed Effects (Entity + Month)
Variable	Ret (1m)	Ret (3m)	Ret (6m)
NPR+	0.000 ***	−0.000	−0.000
	(4.682)	(−0.223)	(−0.503)
NPR−	−0.000 **	−0.000 *	−0.000 **
	(−2.302)	(−1.800)	(−2.000)
Control Variables	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes
Month FE	Yes	Yes	Yes
Cluster by	Firm	Firm	Firm
Observations	13,966	13,817	13,624
Adj. R2 (≈Within)	0.001	−0.000	−0.000
Within R2	0.001	−0.000	−0.000
Overall R2	0.001	−0.000	0.000
Between R2	0.043	−0.002	0.002

Note: This table presents robustness regression results linking insider trading activity to subsequent stock returns under alternative fixed-effects specifications. Panel A reports OLS regressions with industry × month fixed effects, while Panel B reports firm and month fixed effects estimated using a panel data model. The dependent variables are forward stock returns measured over 1-, 3-, and 6-month horizons. ${\mathrm{NPR}}^{+}$ and ${\mathrm{NPR}}^{-}$ represent insider net buying and net selling intensity, respectively. Control variables include firm size (log market capitalization), book-to-market ratio (BM), and idiosyncratic volatility (IVOL, computed over a 12-month window). All regressions include month fixed effects, and standard errors are clustered at the firm level. $t$-statistics are reported in parentheses. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

re-estimates the main regression model under stricter fixed-effects structures. Specifically, two alternative specifications are employed.

The first adds industry × month fixed effects to absorb all sector-specific shocks that vary over time, such as common regulatory changes or industry-wide demand shifts. The model is specified as:

$$RE{T}_{i,t\to t+h}={\beta }_0+{\beta }_1·NP{R}_{i,t}^{+}+{\beta }_2·NP{R}_{i,t}^{-}+{\theta }^{\prime }{X}_{i,t}+{\lambda }_{industry\times month}+{\epsilon }_{i,t,h}$$

(9)

The second includes firm and month fixed effects, which further control for unobserved firm-level characteristics—such as management style, disclosure quality, or governance structure—that remain constant over time:

$$RE{T}_{i,t\to t+h}={\beta }_0+{\beta }_1·NP{R}_{i,t}^{+}+{\beta }_2·NP{R}_{i,t}^{-}+{\theta }^{\prime }{X}_{i,t}+{\lambda }_i+{\lambda }_t+{\epsilon }_{i,t,h}$$

(10)

here, λ_{industry×month} denotes combined industry-month dummies, and λ_i represents firm-specific effects.

Panel A of Table 7 reports results under the industry × month specification. The coefficients on $NP{R}^{+}$ remain positive and significant at the 1% level for short horizons, while $NP{R}^{-}$ remains small and statistically insignificant. This pattern indicates that insider purchases retain predictive power even after accounting for simultaneous industry shocks.

Panel B introduces firm fixed effects. The significance of $NP{R}^{+}$ weakens slightly but remains positive in the 1-month horizon, whereas $NP{R}^{-}$ becomes marginally negative and significant. This suggests that insider selling carries limited but directionally consistent information about near-term price weakness after controlling for all firm-specific characteristics.

Across both models, the estimated coefficients are close in magnitude to those in the baseline regressions, and the explanatory power (Adj. R² ≈ 0.28 to 0.36) remains broadly stable. These findings confirm that the earlier results are not artifacts of omitted heterogeneity or temporal industry effects. The robustness of the insider–return relationship under alternative fixed-effects structures reinforces the reliability of the study’s main conclusions.

4.7. Robustness Tests: Missing-Data Imputation and Election-Year Heterogeneity

Following the missing-data robustness logic in Collier et al. (2025), Panel A of

Table 8. Robustness Tests: Missing-Data Imputation and Election-Year Analysis.

Panel A. Imputation Sample
Variable	Ret (1m)	Ret (3m)	Ret (6m)
NPR+	0.00074	0.00611 **	0.00524 ***
	(0.68)	(2.02)	(2.81)
NPR+ × Tech	−0.00063	−0.00598 **	−0.00527 ***
	(−0.58)	(−1.97)	(−2.83)
NPR+ × Utilities	0.00046	−0.01729 **	−0.01445
	(0.07)	(−2.32)	(−0.53)
NPR-	−0.00005	0.00025000 **	0.00091000 **
	(−1.02)	(2.15)	(2.51)
NPR- × Tech	−0.00024 ***	−0.00073 ***	−0.00186 ***
	(−2.73)	(−4.27)	(−2.87)
NPR- × Utilities	0.00003	−0.00008	−0.00073 **
	(0.56)	(−0.71)	(−2.01)
Control Variables	Yes	Yes	Yes
Month Fixed Effects	Yes	Yes	Yes
Clustered SE	Firm	Firm	Firm
Observations	15,070	15,070	15,070
Adj. R2	0.0123	0.0054	0.0018
Panel B. With Return Seasonality and Election Controls
Variable	Ret (1m)	Ret (3m)	Ret (6m)
NPR+	−0.00055 ***	0.00326 ***	0.00623 ***
	(−2.89)	(10.34)	(4.56)
NPR+ × Tech	0.00063 ***	−0.00321 ***	−0.00612 ***
	(3.32)	(−10.23)	(−4.46)
NPR+ × Utilities	0.00925 ***	−0.01548 ***	0.04706 ***
	(6.05)	(−5.90)	(7.71)
NPR-	0.00005	0.00112 ***	0.0045 **
	(1.21)	(18.33)	(2.02)
NPR- × Tech	−0.00007	−0.00135 ***	−0.00102
	(−0.71)	(−5.09)	(−1.64)
NPR- × Utilities	0.0009 *	−0.00109 ***	−0.00025
	(1.79)	(−11.54)	(−1.01)
Offseason	−0.01008 ***	−0.04780 ***	−0.05438 ***
	(−3.36)	(−7.72)	(−6.04)
PreElection	−0.00758 **	−0.02166 ***	−0.02609 **
	(−2.44)	(−3.20)	(−2.54)
PostElection	−0.00303	0.00370	−0.00291
	(−1.17)	(0.66)	(−0.32)
Control Variables	Yes	Yes	Yes
Year Fixed Effects	Yes	Yes	Yes
Clustered SE	Firm	Firm	Firm
Observations	8428	8360	8252
Adj. R-squared	0.054	0.089	0.124

Notes: This table reports robustness tests of the baseline insider-trading return predictability results using the NPR shares proxy. Panel A (Missing-Data Imputation) re-estimates the models after imputing missing forward returns by month-level medians (with remaining missing values filled by the full-sample median). Panel B estimates the same specifications incorporating the seasonality and the election-related variables. Dependent variables are 1-, 3-, and 6-month forward returns. NPR is decomposed into positive and negative components, with interaction terms for Technology and Utilities; Banking is the omitted baseline sector. All regressions include month fixed effects and firm-clustered standard errors. t-statistics are reported in parentheses on the row immediately below each coefficient. Statistical significance is denoted by *, **, and *** for the 10%, 5%, and 1% levels, respectively.

re-estimates the models after median-based imputation of missing forward returns. The core inference is preserved: positive NPR loads remain meaningful at 3- and 6-month horizons, and technology interactions weaken the purchase signal relative to Banking. The utilities industry again does not display a uniformly stronger effect than the banking industry once the full horizon set is considered.

Building on Waggle and Agrrawal (2018), Panel B now estimates the full sample while controlling for Offseason, PreElection, and PostElection periods. The updated evidence still points to banking as the strongest benchmark for insider-purchase informativeness, especially at the 3- and 6-month horizons where the baseline buy-side NPR effect is most clearly positive. Relative to Banking, the technology industry does not exhibit a consistently stronger buy-side signal and is generally weaker at medium horizons, even after accounting for interaction terms. Utility companies remain mixed across horizons, with signs that change by holding period and no stable outperformance versus the banking sector. Overall, the imputation and election-cycle-control specifications support the same core conclusion: banking purchase signals are the most reliable, technology is comparatively weak, and utilities are intermediate but unstable across models. These controls do not eliminate the insider-trading patterns, suggesting that the documented effects are not solely driven by standard seasonal or election-cycle variation.

4.8. Economic Significance and Portfolio Efficiency of Insider Signals

The compact portfolio-sorting results indicate that the insider-trading signal contains economically meaningful information at the monthly horizon. The results are presented in

Table 9. Portfolio Sorts Based on Insider Trading Signals.

Signal	EW Q5-Q1 Diff	EW T-Stat	VW Q5-Q1 Diff	VW T-Stat	EW Annualized	VW Annualized
NPR	−0.60% *	(−1.90)	−0.78% **	(−2.04)	−7.17%	−9.41%
NPR+	−0.09%	(−0.34)	−0.14%	(−0.36)	−1.02%	−1.69%
NPR-	0.59% *	(1.90)	0.71% *	(1.82)	7.05%	8.46%

Note: This table reports portfolio-sort results linking insider trading activity to subsequent stock returns. In each month, firms are sorted into quintiles (‘Q1’ to ‘Q5’) by insider trading signals, including ‘NPR’, ‘NPR+’, and ‘NPR-’, and the reported spread is the top-minus-bottom portfolio return (‘Q5-Q1’) for ‘NPR’ and ‘NPR+’, and the low-minus-high spread for ‘NPR-’ (to reflect stronger sell pressure in more negative values). Returns are measured over the next month. Results are shown for both equal-weighted (‘EW’) and value-weighted (‘VW’) portfolios. Reported statistics include mean monthly spread return, t-statistic (in parentheses), and annualized return. ‘NPR+’ and ‘NPR-’ denote insider net buying and net selling intensity, respectively. Significance markers ‘**’, and ‘*’ indicate the 5%, and 10% levels, respectively.

. In particular, the long–short spread between the top and bottom quintiles of NPR is negative and statistically significant, with value-weighted returns of approximately −0.78% per month (about −9.4% annualized), implying that firms with high insider selling pressure subsequently underperform those with low pressure. Further decomposing the signal, NPR+ (buy-side intensity) exhibits weak, statistically insignificant spreads, suggesting limited standalone predictive power after sorting. In contrast, NPR− (sell pressure) generates positive spreads when constructed as low-minus-high portfolios, consistent with the notion that stronger insider selling predicts lower future returns.

In economic terms, these spread magnitudes are non-trivial. Annualized gross returns in the 7–9% range suggest potential relevance to practical allocation decisions, particularly within a wealth-maximization framework, as emphasized by DiLellio and Stanley (2011), who discuss ETF-based strategy design to enhance client wealth. From a portfolio-efficiency perspective, the critical issue is whether these returns simply proxy for known risk premia or reflect incremental information. If the spreads remain after accounting for transaction costs and standard asset-pricing factors, the evidence supports the view that insider-trading signals convey information not fully captured by conventional benchmarks and therefore may improve investor welfare. This portfolio evidence complements, rather than replaces, the regression results. The regression framework is designed to separate buy-side and sell-side components across industries, whereas the portfolio sorts summarize the overall return spread associated with the insider-trading signal at the monthly horizon.

5. Conclusions

This study utilizes insider trading data from 2005 to 2025 to investigate whether insider activity can predict future stock returns across technology, banking, and utility firms. The patterns in the data show clear differences across these industries. Insider purchases consistently have the strongest predictive power in banking firms, where closer supervision and more standardized disclosures limit trades unrelated to firm fundamentals. In contrast, the signal from insider buying is much weaker in technology firms, likely because frequent news releases and analyst coverage narrow the gap between insiders and outside investors. Utilities fall in the middle, with insider purchases showing some predictive content but not at the level observed in the banking sector.

The results also show that insider buying is generally more informative than insider selling. Many sales transactions are motivated by liquidity needs, diversification, or tax considerations, which makes their connection to private information less clear. This buy–sell difference is not the same across sectors. In banking and utilities, the gap between the informativeness of purchases and sales is more visible, while technology firms show much less separation between the two types of trades. These patterns suggest that industry structure and information environments shape both the strength of insider signals and the meaning of different trade types. The main findings remain stable when using alternative measures of insider activity, risk-adjusted returns, and different fixed-effects specifications.

This study contributes to the insider trading literature in several ways. First, it links insider predictability to the information environments of different industries, showing that insider trades do not carry the same meaning across sectors. This addresses a gap in prior work, which often treats insiders as a single group and overlooks how transparency and supervision shape the strength of insider signals. Second, the results show that the familiar buy–sell asymmetry is not a universal pattern but one that depends on industry conditions, particularly the level of monitoring and the stability of public disclosures. The paper also provides early evidence on how the recent changes to Rule 10b5-1 are associated with differences in insider trading patterns, particularly a decline in the informativeness of sell-side trades. These findings provide a clearer understanding of how regulatory and information environments interact, and they may help inform future studies on how policy changes influence the flow of private information into financial markets.

Like most empirical studies, this paper has several limitations that point to opportunities for future research. The dataset does not allow for a clean separation between opportunistic insider trades and routine transactions arising from diversification needs or preset trading plans, which may influence the interpretation of sell-side activity. Additionally, industry classifications rely on Bloomberg sector codes, and a more detailed or multidimensional approach may more accurately capture differences in information environments. Future work could incorporate textual analysis of earnings calls or management forecasts to better measure how firms disclose information and how quickly the market absorbs it. Another promising direction is to examine how insider compensation arrangements—such as stock option grants, vesting schedules, or equity-based incentives—affect trading behavior and the informativeness of insider signals. These extensions would help build a more complete understanding of how private information is generated, used, and transmitted across different market settings.