Signal performance

All our return metrics follow the same convention. They are deliberately conservative, it's the return a retail investor can actually capture, not the insider's own return.

Local regulations (MAR 596/2014 in the EU for AMF/BaFin, SEC Form 4 in the US, SIX SER in Switzerland) require executives to file their transactions within a short window (3 business days in the EU, 2 in the US). But do they comply? On our sample of 221,095 filings:

Conclusion: the majority (56%) comply with MAR. The median is 2.5 days, meaning that when you see a signal on the site, the transaction already occurred 2.5 days ago on average. The market has had time to price the information.

Key question for a retail investor: how much of the stock's total return happens between the insider trade and the regulatory filing (leak) vs. after (exploitable by you)? We measured the Yahoo price at transactionDate, at pubDate+1, and at tx+90d for 232,123 trades.

We tested 6 strategies on the same universe, with the same rules:

• Top-20 portfolio · we buy the 20 best signals from the previous month, sorted by score.
• Monthly rebalancing · old positions out, new ones in.
• 3-month hold per position (matches the T+90 backtest window).
• Brokerage fees 1% round-trip (French retail price).
• Entry at pubDate+1 (the day after the regulatory filing).

Reading the table: the Signals column shows how many trades matched the filter over the full period. Sharpe = annualised return/risk ratio.Max DD = worst cumulative loss during the simulation.Beats CAC = % of months the strategy outperformed the CAC 40.

T+90 ↔ CAGR conversion. T+90 returns are per-trade, 90-day holds; CAGR is the compounded annual return of a portfolio that recycles the signal continuously. Example: a +2.8 % average T+90 compounded over ~4 cycles per year ≈ +12 % CAGR (formula: (1+0.028)^365/90 − 1). The two numbers are consistent; they answer two different questions ("how much per trade?" vs "how much per portfolio-year?"). The published win rate is the % of signals closing in profit at T+90.

Beyond the Sharpe / CAGR pair, here are the six risk metrics we now publish on the production V14e EU_strict strategy. They answer the questions a real investor actually asks: how much can I lose, for how long, and whether my upside volatility pays for my downside.

Definitions. Annualised vol = std-dev of monthly returns × √12. Max DD = worst peak-to-trough loss on the monthly equity curve (not daily, see caveat below). Time-under-water = consecutive months below a prior peak. Calmar = CAGR / |Max DD|, more meaningful than Sharpe for tail-risk strategies. Sortino = return / downside deviation; distinguishes upside vol (welcome) from downside vol (suffered).

Hit ratio by sector

Win rate and average T+90 return broken down by Company.sectorTag, on the Sigma v1.0 historical discovery cohort (pre-V14e). Highlights where alpha concentrated and surfaces sector concentrations capped at 30 % by the risk-management caps.

No, you do NOT need to hold all 585 companiesto extract value from the system. It's all about selection and discipline. Here is our recommendation for an investor with 20 positions maximum and capital between €10,000 and €100,000.

• Backtest ≠ future. Our +8.1% are historical over 4 years. Markets and insider behaviours evolve. The future can be better… or worse.
• Survivorship bias. Our database contains companies still listed. Delisted or bankrupt ones are not counted, which could overstate real performance by 0.5 to 1 point.
• Slippage not modelled. On illiquid small-caps (typically EU and emerging mid-caps), the bid-ask spread can reach 1-3%. Our backtests assume execution at closing price. In practice, you will pay the ask and sell at the bid → subtract an extra -1 to -2% from returns.
• Limiting liquidity. Some small-caps have daily volume < €100k. Putting €5k into a micro-cap signal can move the market. Sigma has no value on these stocks for investors > €500k.
• Short backtest horizon. 4 years is insufficient to judge an investment system. Academics typically require 10-15 years. We continuously enrich the database to reach 10 years by 2029.
• Favourable CAC period.2021-2025 was particularly bullish for the CAC 40 (ATH in 2024). In a bear cycle, our low-drawdown signals could catch up with the benchmark, but we don't yet have the data to prove it.
• Not investment advice. This site is an information tool. It is neither a robo-advisor nor a fund manager. Your decisions are yours; we are not responsible for losses.

• Scoring v3 deployed (April 2026). Methodology overhaul after audit: cluster detection is now strictly directional, added insider track record (Bayesian shrinkage), DCA bonus and analyst-contrarian signal. Staleness penalty removed from the stored score (applied at display time only). Weight redistributed from public information (fundamentals, composite) toward insider-specific signals.
• Bayesian shrinkage on recommendations.Under-sampled buckets (role × size) are now pulled toward the global mean with a size-adaptive prior (k=30 if n<20, k=15 if 20≤n<200, k=5 if n≥200). No more structural over-ranking of low-n buckets on noise. The recency half-life was raised from 21 to 45 days (aligned with the 90-day hold).
• New "Sweet spot" bucket (€300M-€1B). Split from Mid to isolate the zone where the grid search found alpha concentrates (liquid, under-covered). Taxonomy now in 6 sizes: Micro / Small / Sweet / Mid / Large / Mega.
• Real retail returns stored. Each backtest now containsreturnFromPub30d/90d/365d (entry at pubDate+1) in addition to insider-view returns (transactionDate). Future backtests will be more honest by default.
• Automated monthly backtest. Every Sunday at 5h UTC, we recompute 300 missing backtests. The system becomes more accurate each week.
• Historical expansion. Back-filling filings 2015-2020 in progress (currently available: 1922 → 2026).
• Sell signals. Our engine supports BUY and SELL, useful for position management on a held stock.