Can You Profit by Following Corporate Insiders? A New Study Says Yes — With Caveats

Jan Schroeder and Alexander Krause, authors of the study “Following Insiders to Outperform the Market,” published in the June 2026 issue of The Journal of Investing, revisit a question that has interested both academics and investors for decades: Does mimicking the stock purchases of corporate insiders generate market-beating returns? The authors bring a new portfolio-construction lens to the uestion — along with a few reasons for skepticism.

What the Authors Examined

The study analyzed 1.8 million SEC Form 4 filings from 2011 through 2023, covering insider transactions at 9,166 companies. Unlike most academic studies in this space, which stop at calculating abnormal returns around a disclosure event, the authors went a step further: they built an actual rule-based trading algorithm, simulated it with realistic constraints, and measured its performance against the S&P 500 using 14 separate risk and return metrics — not just cumulative abnormal return (CAR).

A few design choices stand out:

· Purchases only, not sales. The authors excluded insider sales, reasoning that purchases reflect a single, voluntary signal (a bullish view), while sales are muddied by liquidity needs, diversification, tax planning, and pre-scheduled 10b5-1 plans. They also note the practical difficulty of simulating short positions given borrowing costs and share availability.

· Realistic execution assumptions. Positions are entered at the next day’s market open — not the same-day close — since roughly 75% of Form 4 filings are disclosed after 4:00 p.m. Using same-day closing prices, as some earlier studies did, would have introduced look-ahead bias.

· Risk controls and costs. Each position is held a maximum of five trading days or exited early on a 15% stop-loss, with 0.1% transaction costs charged on entry and exit. Capital is reset to $100,000 at the start of each year so that compounding doesn’t distort the return metrics.

· A Monte Carlo search for filter rules. Rather than hand-picking filters that align with theory, the authors ran Monte Carlo simulations across 1,685 firm- and filing-level features (sector, insider role, financial ratios, growth metrics, and more) using only in-sample data (2011–2019) to identify the filter combination with the strongest five-day CAR. That combination — “Strategy A” — was then tested out-of-sample (2020–2023) to check whether it held up on data it hadn’t seen.

For comparison, they also tested an unfiltered baseline (Strategy B, all insider purchases) and three sector-specific strategies: technology (C), health care (D), and utilities (E).

Key Findings

Event-study results. Strategy A’s five-day average CAR was 3.42% (median 2.16%), well above the unfiltered baseline’s 1.19% (median 0.37%). Sector-only filters did better than the baseline but not as well as the multi-factor Strategy A — health care delivered a 2.97% five-day CAR and technology 1.68%, while utilities barely moved the needle (0.25%).

Backtested performance. Over the full 2011–2023 period, Strategy A produced a 708% cumulative return versus 377% for the S&P 500, a 17.4% CAGR versus 12.8% for the index, a Sharpe ratio of 1.88, and a Sortino ratio of 2.92. It outperformed the S&P 500 in 100 of 156 months (64%). Importantly, the strategy held up out-of-sample (2020–2023), which is the part of this study I find most credible — the filter rules were derived strictly from 2011–2019 data and not re-optimized on the test period.

The trade-off is real. Strategy A didn’t win across the board. While it produced a shorter longest drawdown period com­pared to the S&P 500 (116 days vs. 259 days) and all other strategies, it underperformed the S&P 500 on three risk metrics: maximum drawdown (–38.7% vs. –33.9%), annualized volatility (29.3% vs. 14.6%), and 95% Value at Risk (VaR) (–2.03% vs. –1.54%). In other words, the higher returns came with materially more risk — a point the authors are appropriately upfront about, and one advisors should weigh heavily before reading the headline Sharpe ratio in isolation.

Trade characteristics. Strategy A averaged about 400 trades a year across a maximum of 12 simultaneous positions, with a 55.7% win rate and a 1.45 profit factor — the best of any strategy tested, including the unfiltered baseline and the sector-specific cuts. Strategy A also benefited from relatively low transaction costs. Over the 13-year back tested period, This lower trade vol­ume minimized transaction costs, further enhancing Strategy A’s overall perfor­mance advantage.

What Exactly Did Strategy A Involve?

It is worth being precise about what “Strategy A” actually is, since the paper leans heavily on its results without spelling out the rule itself in plain terms. Strategy A is the single best-performing filter combination to emerge from the Monte Carlo search across the dataset’s 1,685 features — covering both Form 4 filing details (insider relationship to the company, purchase volume, disclosure timing) and company-level fundamentals (sector, market cap, financial statement items, ratios, growth metrics). On each simulation run, the algorithm randomly drew a subset of these features; for categorical variables like sector or insider role it randomly selected a category, and for numerical variables like revenue it randomly defined a range. Each randomly generated rule set was then applied only to in-sample insider purchases (2011–2019) and scored on the resulting median five-day CAR, with the best-scoring combination becoming Strategy A. That combination was subsequently tested out-of-sample on 2020–2023 data.

Here’s the catch: the authors describe Strategy A only as “a combination of sector-specific, financial metric, and insider-related filters.” They never disclose which sector(s), which specific financial-metric thresholds, or which insider-relationship category actually define the rule. That’s a real gap for anyone trying to evaluate or replicate it — we know how the rule was found, not what the rule says.

What we can infer is the shape of the resulting portfolio from the company-characteristic tables the authors do provide (Exhibits 10 and 11). Relative to the unfiltered universe (Strategy B), the transactions matching Strategy A’s rules:

1. Came from a much narrower slice of the data. Only 12,903 of the original 180,277 insider purchases — about 7% — met the filter criteria.

2. Skewed toward smaller, less profitable companies. Median revenue for Strategy A’s matched transactions was $105 million versus $226 million for the unfiltered universe, and median net income was −$5.3 million versus +$7.5 million. Median open price was $30.94, modestly above Strategy B’s $15.81.

3. Traded relatively infrequently. Strategy A generated roughly 5,200 total trades over 13 years — about 400 a year — across 1,716 unique tickers, versus roughly 77,000 trades across 7,239 tickers for the unfiltered baseline.

So while the paper frames Strategy A as a rigorously derived rule, what’s actually disclosed is the search process and the statistical fingerprint of the companies it selects — not the rule’s content. That fingerprint, smaller and less profitable names, is consistent with the size/illiquidity-tilt concern raised below.

Why I’d Stay Cautious About the Headline Numbers