A correct forecast with wrong size is still a risk management failure. Expected value and Kelly tell you how much edge deserves; portfolio construction tells you how much cluster allows. This chapter turns both into a bankroll policy you repeat when caps, curves, and depth disagree across venues.
Forecasters love debating probability; survivors obsess over how much. Sizing is where calibration meets bankroll reality—especially when Kelly wants more than the book or the regulator allows.
What bankroll means
Bankroll is money you can lose without breaking life obligations. Open risk is worst-case loss on open positions—not paper gains. Ruin is falling below the minimum where disciplined sizing still makes sense. Binary contracts cap loss per share at premium paid, but correlation does not cap across five politics legs.
Frameworks in one paragraph
Flat units are simple but ignore edge quality. Fixed fractional risk (one to two percent per idea) survives streaks. Kelly maximizes long-run growth in theory but needs honest probabilities and independence you rarely have. Vol targeting scales down when recent swings rise—needs history from journaling. Platform and liquidity caps are hard ceilings that trump math.
Many hobbyists land on half Kelly capped at two percent per binary, tighter during election clusters.
The sizing pipeline
State probability and executable price. Compute edge after fees. Compute frictionless Kelly, then apply a quarter to half fraction. Haircut for correlated open bets. Apply venue maximums and level-one depth. Log planned versus actual. If edge is zero, size is zero—unless you explicitly label play-money tuition.
Worked cap example
Bankroll $20,000. YES at 36¢, your probability 48%. Full Kelly might suggest nearly 19%—about $3,750. Half-Kelly halves that. If you already have 40% in the same election tree, halve again toward $900. If level-one ask is only $600, you take $600. Math screamed big; book and theme whispered small.
On a capped venue, $850 per contract might bind before Kelly even if your edge is real—plus winner fees shrink the win branch. Regulatory cap is not “conservative Kelly.”
Bankroll tiers
Learning accounts under a thousand dollars might risk two percent or play-money only. Serious hobbyists from one to twenty-five thousand often use one to three percent half-Kelly with fifteen percent correlated bucket limits. Larger side-income books add vol scaling and hedges. Moving signal from a capped site to a primary venue is a new mental account, not the same bankroll line.
Correlation and arb sizing
YES president plus YES Senate for the same party is one bucket at half combined Kelly unless joint modeling says otherwise. Arbitrage looks small risk but ties capital on both sides—size to the worse leg and to unwind stress, not to theoretical locked cents.
Manifold mana wins must not scale USD without Brier evidence on real fills.
Adds, averages, and drawdown rules
Add to winners when new information raises probability, liquidity allows, and cluster budget has room—not when price alone spiked. Tranche entries respect thin books: three clips of four hundred dollars beat one market sweep that walks the curve. Averaging down when thesis is impaired is forbidden; exit rules belong in the stop chapter.
At minus ten percent from peak bankroll, halve unit size until recovery. At minus twenty percent, pause new risk and audit the journal. Three consecutive oversize trades force fractional Kelly for the next twenty tickets.
Common errors
Sizing from conviction tone, ignoring fees inside probability, treating arb as zero risk with one stuck leg, using the same percent on ninety-cent and ten-cent contracts without translating to dollars at risk, revenge doubling after loss, and labeling a fifth correlated politics leg “diversification.”
Units versus dollars
One percent of bankroll on a 90¢ YES is not the same risk story as one percent on a 10¢ YES—the latter has more upside and more ways to be wrong about tail events. Translate size to premium dollars at risk and to profit if right so you see skew.
Learning-phase policy
Until twenty resolved logged trades, halve every Kelly output again. Calibration beats bravado. Manifold and capped venues are for process practice; USD size waits for Brier evidence.
Interaction with stops and bias
Oversize makes thesis stops unaffordable emotionally—that is how loss aversion wins. Right size makes holding through noise possible. Caps are bias controls, not punishments.
Tranche example in prose
Edge is nine cents but level-one depth is only four hundred dollars at 52¢. Plan three tranches of four hundred dollars: first at touch, second on a limit one cent better, third only if probability unchanged after the debate. Total twelve hundred dollars stays under three percent of a forty-thousand-dollar bankroll and respects the book without one obese market order.
Key ideas
Pipeline: edge → fractional Kelly → correlation → venue → depth. Bankroll policy beats heroic tickets. Drawdown rules protect future you.
What comes next
Size and fill mean little without a coherent exit policy.
Passive versus active size
Passive holds can use larger fractional Kelly when correlation is low and horizon is long—variance is expected, not a surprise. Momentum and contrarian clips should default smaller because clustering and fees compound. Arb size is bounded by worse-leg depth and float, not by Kelly on a fictional risk-free bet. Label the strategy in the journal so you do not size a scalp like a six-month macro hold.
Written bankroll policy
Put your pipeline on one page: bankroll definition, fractional Kelly choice, cluster cap, single-leg cap, drawdown triggers, and venue ceiling. Reference that page in every journal row. When you break the page, you are not “trading bigger because edge”—you are out of policy. Rewrite the page monthly with data, not with feelings.
Sizing is the bridge between forecasting chapters and risk management in this module. Get it wrong and correct probabilities still lose the account; get it right and modest edges compound.
When in doubt, smaller and more frequent review beats heroic size and rare review. The journal proves which choice matched your true calibration.
Next: Stop-Losses in Prediction Markets (When They Work)