Risk Management and Bankroll: Surviving Variance Long Enough to Win

Most traders who quit prediction markets don't quit because they couldn't find an edge. They quit because they sized one good idea too big, took a normal losing streak personally, and blew up before their edge had time to show. Finding +EV signals is the glamorous half of trading. Surviving the variance around that edge is the half that actually determines whether you're still here in a year. This guide is about that second half: how to define a bankroll, size positions, cap concentration, cut losers, diversify, and keep your head after a bad week — and how to run Predite's Kelly Calculator, Stop-Loss, and Trade Journal together as one risk system.

What a Bankroll Actually Is

Your bankroll is the total pool of money you have allocated to prediction-market trading — and nothing else. It is not your net worth. It is not your savings. It is not money you'll need for rent in three months. It's the amount you've decided you can expose to a high-variance activity, mentally written off as "at risk," and will manage as a single number.

Defining it cleanly matters because every sizing decision downstream is a percentage of this number. If your bankroll is fuzzy — "somewhere between $2,000 and the $8,000 in my checking account I could tap if a trade looks really good" — then your position sizes are fuzzy too, and fuzzy sizing is how people end up 40% deep in one market.

A few rules for setting it:

Position Sizing with Fractional Kelly

Once you have a bankroll and a +EV opportunity, the question is *how much*. The Kelly Criterion answers it precisely. For a binary prediction-market contract, where you buy YES at price P and your subjective probability is p, the optimal fraction simplifies to:

f\* = (p − P) / (1 − P)

That's full Kelly. Our companion guide, *The Kelly Criterion: Sizing for Compounding*, derives this and works through the math; here we'll focus on using it safely.

Take a concrete trade. You estimate 58% on a market priced at $0.45 for YES. Full Kelly is (0.58 − 0.45) / (1 − 0.45) = 0.13 / 0.55 ≈ 0.236, or 23.6% of bankroll. On a $5,000 bankroll that's $1,180 on a single market that you think is only mildly underpriced. That should feel like too much — because it is.

Full Kelly assumes your probability estimate is *exactly* right, that you can rebalance instantly, and that you can emotionally survive 50%+ drawdowns. None of those hold for real humans with approximate estimates. So you bet a fraction of what Kelly says:

The Predite Kelly Calculator (Dashboard → Kelly Calculator, available on Pro and Bot plans) does this live. You enter bankroll, your probability, and the market price; it returns your recommended USD size, the edge in percentage points, expected value per trade, and the expected geometric growth rate. The fraction slider defaults to 0.25, and there's a what-if table showing 25/50/75/100% Kelly side by side so you can *see* how fast the suggested size — and the implied drawdown risk — climbs as you get greedier.

Two safety features are baked in. If your probability is at or below the market price, the calculator returns zero and tells you not to bet — there's no edge to size. And there's a hard 25% absolute cap: even if Kelly mathematically suggests more, the tool clamps the recommendation to a quarter of your bankroll and flags it, as a fail-safe against an over-optimistic estimate. Treat that cap as a ceiling you rarely approach, not a target.

Caps: Per Market and Per Correlated Cluster

Kelly sizes one bet in isolation. Real portfolios need two more limits layered on top.

Max per single market. Regardless of what Kelly says, cap any one market at 25% of bankroll — and in practice, with quarter Kelly, you'll almost never get near it. The point is that no single unexpected resolution should be able to cut your stack by more than a quarter. For illiquid markets (daily volume of $10k or less), tighten that to 5–10%, because the price you see is not the price you'll get when you try to exit in size.

Max per correlated cluster. This is the rule most traders miss, and the one that quietly causes the worst blowups. If you hold three positions that all resolve on the *same underlying event*, you don't have three bets — you have one bet in a trench coat. Consider:

Those are nominally three markets across two categories, but a single hawkish surprise tanks all three at once. If you'd sized each at 8% thinking you were diversified, you're actually 24% exposed to one Fed decision. Cap total exposure to any correlated cluster at 50% of bankroll, and ideally keep it well under that. When in doubt, ask: "If the single most important fact resolved against me tomorrow, how much of my bankroll moves?" That number is your true position size.

When you have several +EV trades at once, don't naively add their Kelly fractions — that over-leverages you. Sum them, and if the total exceeds your exposure ceiling (say 50%), scale every position down proportionally. Three trades each suggesting 20% Kelly = 60% total → scale all three to ~17% so the sum lands at 50%.

Stop-Loss Discipline

Prediction markets let you exit before resolution, which means a losing position is a *choice* to keep holding, not an inevitability. A stop-loss removes the choice — and removes your emotions — by pre-committing to an exit at a defined price.

Why pre-commit? Because the moment a position is underwater is exactly the moment your judgment is worst. "It'll come back" is the thought that turns a planned 20% loss into a 100% one. Deciding the exit *before* you're in pain is the whole point.

Polymarket has no native stop order, so Predite runs stops client-side: a cron job checks your open positions' prices every minute and fires the close when a trigger hits. There are three trigger kinds:

Stops attach to paper positions and to bot positions, and they're idempotent — once triggered, a stop can't accidentally fire twice. A practical default: set a stop-loss and a take-profit on every position the moment you open it, sized so your downside is capped at roughly 1–2% of *total bankroll* per trade. A $295 position with a stop that caps the loss near $70 risks ~1.4% of a $5,000 bankroll — survivable dozens of times in a row.

One nuance specific to these markets: don't set stops so tight that normal noise knocks you out. A market that swings $0.04 on a single tweet will stop you out repeatedly if your trigger is $0.03 away. Place stops outside the market's typical chop, at the level where your *thesis* is genuinely broken — not where a routine wiggle reaches.

Diversification Across Categories and Platforms

Diversification doesn't increase your edge; it reduces the variance around it, which is what keeps you solvent long enough for the edge to compound. The goal is a book of bets whose outcomes are as independent as possible.

Variance vs Edge Over a Sample

Here's the most important mental model in this guide: a positive edge guarantees nothing over any short run. Edge is what happens on average over a large sample. Variance is what happens to you this week.

Suppose you genuinely have a real 55% win rate on roughly even-money bets — a strong, durable edge. Over 20 trades, basic binomial math says you'll lose money more than a quarter of the time through sheer luck. Losing streaks of 5, 6, even 7 in a row are *normal* and *expected* at a 55% win rate; across a few hundred trades you should anticipate several of them. None of that means your edge is gone.

This is why the rules above are non-negotiable. Quarter Kelly, the 25% market cap, and per-trade stops near 1–2% of bankroll exist precisely so a normal unlucky streak is a dent, not a death. Run the arithmetic: ten consecutive losses at 1.5% of bankroll each is about a 14% drawdown — uncomfortable, fully recoverable. Ten consecutive losses at full-Kelly sizing of 20%+ each is a wipeout from which no edge can save you.

The corollary: judge your strategy by your sample, not by your last trade. You need roughly 100+ resolved trades before your realized win rate says anything trustworthy about your true edge. Until then, stay at quarter Kelly or below and resist the urge to conclude anything from a hot or cold run of ten.

Avoiding Tilt After Losing Streaks

Tilt is the gap between the rules you wrote down when calm and the trades you actually place when hurting. It is the mechanism by which good strategies produce bad results. After a losing streak, three predictable distortions kick in:

1. Revenge sizing — betting bigger to win it back faster. This is *anti-Kelly*: you're increasing position size exactly when your bankroll, and often your judgment, has shrunk. Lethal over a long enough series.
2. Edge inflation — convincing yourself a marginal trade is a great one because you *need* a win. Your conviction quietly rewrites your probability estimate upward.
3. Stop-jumping — canceling or widening a stop as the price approaches it, because taking the loss feels unbearable in the moment.

Concrete defenses that work:

Running the Three Tools Together

Risk management isn't one feature; it's a loop. On Predite, the Kelly Calculator, Stop-Loss, and Trade Journal (all on Pro and Bot plans) form a closed cycle: size correctly, protect automatically, then learn from the record so your next sizing is sharper.

A repeatable per-trade workflow:

1. Find a +EV signal with the EV Scanner and apply a conservative haircut to the AI probability (if it says 56%, treat it as ~54%) — models are calibrated but never perfect, and humility here is free insurance.
2. Size it in the Kelly Calculator. Enter your current bankroll, your haircut probability, and the market price. Use the 0.25 fraction default. Sanity-check the output against your per-market and per-cluster caps before you commit a dollar.
3. Place the trade, then immediately set Stop-Loss and take-profit. Cap the downside near 1–2% of total bankroll. Add a time_stop if your thesis has an expiry. Do this *now*, while you're calm — not later, while you're hoping.
4. Log it in the Trade Journal. Record your entry, your estimated probability, your thesis, and your size. This is the raw material for everything you'll learn.
5. Review the analytics regularly. The Journal's analytics surface the patterns that matter: your realized win rate versus your *predicted* win rate (the direct test of whether your edge is real), which categories you're actually profitable in, and whether your results decay when you size up. If you predict 58% winners but realize 52%, your model is mis-calibrated — trade smaller and tighten your estimates until the two numbers converge.

That last loop is what compounds. Sizing keeps you alive through variance; stops keep any single trade from becoming a catastrophe; the journal turns a pile of outcomes into a measured, improving edge. Skip the journal and you're flying blind — repeating the same mis-sized mistakes with no feedback to correct them.

The traders who last aren't the ones who never lose. They're the ones whose losses are small, bounded, and informative. Open the Kelly Calculator on your next signal, set a stop the moment you're filled, and log the trade — do that for a hundred trades and you'll have something most people in this market never build: hard evidence of whether your edge is real, and a bankroll still standing to press it.