Anti-tilt Is the Only Alpha Nobody Wants to Hear About

We have spent two years looking at anonymised user PnL histories on our platform. The biggest single predictor of a blown account is not strategy choice, not leverage, not exchange, not asset class. It is the time between a losing trade closing and the next trade opening. Specifically: the window from zero to forty minutes after a stop-out. The revenge-trade window.

• The 40-minute post-stop window is 8% of trades by count and about 22% of lifetime drawdown.
• Those trades lose more often (34% vs 48%), at larger size (1.9x) and higher leverage (1.4x).
• A genuine strategy edge can be entirely consumed by sixteen such trades a month.
• "Just don't do it" fails because the physiological state presents itself as clarity, not as tilt.
• The only reliable fix is structural: a control that acts without asking you at the moment you are worst placed to answer.

The revenge-trade signature

Here is what it looks like in the data. A user's win rate, averaged across all trades, is 48%. The win rate on trades opened within 40 minutes of a stop-out is 34%. The average position size on those revenge trades is 1.9× the user's own median position size. Their average leverage is 1.4× their own median leverage.

So not only is the trade more likely to lose: it loses larger, and it loses with more leverage. The 40-minute post-stop window accounts for roughly 22% of lifetime drawdown for the typical active user, while representing only 8% of trades by count.

Let that number sit for a second. One fifth of the losses come from a tiny slice of the trade log, concentrated in the worst 40 minutes of the trader's day.

Why strategy alpha cannot save you from this

Assume your strategy has a genuine edge worth 15 basis points per trade after costs. Assume you take 200 trades per month. That is 3,000 bps of gross alpha, or 30% of starting capital, before any drawdown management.

Now assume 8% of those trades (sixteen per month) are revenge trades sized at 1.9× and levered at 1.4×. Historical data says their net expected return is around −80 bps each, after costs. Sixteen trades × −80 bps × 1.9 size multiplier = −2,430 bps of drag.

Your 3,000-bps edge nets out to 570 bps of actual return. Your strategy was right. Your trader was wrong at the worst possible moments. The strategy got no credit for being right, because the execution of the strategy happened through a person who was in physiological stress for 40 minutes after every loss.

Edge at the chart-pattern level is worth almost nothing if you donate it back during autonomic-nervous-system events.

Why "just stop doing it" does not work

Trading literature has been telling people to "not revenge trade" for fifty years. It does not work. The reason it does not work is that by the time you are aware of the impulse, you have already been dosed by your own adrenaline and the urge to act is being interpreted by your brain as certainty. You do not feel tilted. You feel clear. You feel like now is obviously the right time. That feeling is the tilt.

You cannot will your way out of a physiological state by labelling it. You can only design around it.

What anti-tilt actually looks like, as a product feature

We built four things specifically around the 40-minute window.

1. Post-stop cooldown

After any trade closes in the red on an account above a user-defined loss threshold (typical: 0.5R), the platform enforces a 40-minute cooldown on that coin. No new positions on the same asset during that window. You can trade other coins. You can modify existing positions. You cannot re-enter the instrument that just hurt you.

Users turn this on in onboarding. 71% of users who enable it leave it on permanently. Of the ones who turn it off, almost all do so within the first week: the ones who keep it past week one keep it forever.

2. Size-ratchet guard

If the user attempts to open a new position within 2 hours of a loss, and that new position is larger than their trailing 30-day median size, the platform requires a confirm-by-typing step. "Type REVENGE to confirm." Most people who see the dialog close it and do not come back for twenty minutes. That twenty minutes is usually enough.

3. Mood log, tied to trades

The journal prompts for a single-tap mood entry at the close of every trade: sharp / flat / tilted / unknown. The entries are correlated against subsequent trade outcomes in the analytics view. Users can see, on their own data, that their "tilted" entries have a 31% win rate versus 52% on "sharp" entries. This is not us telling them. This is their own trade log, with their own labels, in their own numbers.

4. Playbook enforcement

Users define their own trading rules in the Playbook tab. "No trades in the first 5 minutes after a loss." "Max 2 losses per day, then flat." "No adds to losing positions." These rules are checked at order-placement time and block the action if they are violated. A rule written by a calm person binds the actions of the tilted person fifteen minutes later. This is the whole trick.

Why nobody wants to hear about this

Anti-tilt features are unsexy. They do not appear in a backtest. They do not photograph well in screenshots. They cost the user nothing on their best day and save their account on their worst day. They are the trading equivalent of seat belts: boring, mandatory, life-saving, impossible to market with.

But they are the single largest adjustable-knob improvement in realised returns that we have ever measured. Not chart-reading. Not signal choice. Not leverage management. Revenge-trade prevention.

// questions or corrections · [email protected] · more essays · /blog