The Kelly Criterion Applied to Tariff Prediction Markets

You've done your homework: analyzed USTR dockets, tracked Congressional signals, and identified that the market is pricing China ETR 20-25% bucket at $0.45 when your model says it should be $0.68. You have edge. The question: how much should you bet?

Bet too little, you leave money on the table. Bet too much, you risk blowing up on a single bad outcome—even when you're right on average. The Kelly Criterion solves this by calculating the mathematically optimal bet size that maximizes long-term growth rate while minimizing ruin probability.

For tariff prediction markets where you have genuine information advantages (policy analysis skills, faster data processing, better USTR docket tracking), Kelly sizing can mean the difference between slow, steady compounding and catastrophic drawdowns. This guide explains how to apply Kelly to tariff trades, when to use fractional Kelly, and why position sizing matters more than win rate.

What is the Kelly Criterion?

The Kelly Criterion, developed by John Kelly at Bell Labs in 1956, answers one question: what fraction of your bankroll should you wager on a bet with a given edge and payoff?

The Formula

For binary outcomes (YES/NO markets):

f* = (bp - q) / b

Where:

Example: You estimate 60% chance of winning, offered 2:1 odds.

f* = (2 × 0.60 - 0.40) / 2
f* = (1.20 - 0.40) / 2
f* = 0.40 (bet 40% of bankroll)

For Bucketed Scalar Markets

Tariff markets use bucketed scalars (e.g., "20-25% ETR"), not binary YES/NO. The formula adjusts:

f* = (p × V - C) / V

Where:

Example: China 20-25% bucket costs $0.45. You estimate 68% chance it wins.

f* = (0.68 × $1.00 - $0.45) / $1.00
f* = ($0.68 - $0.45) / $1.00
f* = 0.23 (bet 23% of bankroll)

Why Kelly Works: Maximizing Growth Rate

Kelly sizing mathematically maximizes the geometric mean of your bankroll over infinite trials. This translates to:

1. Fastest Long-Term Growth: No other bet sizing strategy compounds faster over time.

2. Zero Ruin Probability: As long as you have genuine edge (p > C), Kelly never risks total bankroll loss.

3. Captures Edge Efficiently: Higher edge → larger bet. Lower edge → smaller bet. Automatically scales risk to opportunity.

The Math Behind Growth Rate

Suppose you make the same bet 100 times with 60% win rate at 2:1 odds. Kelly says bet 20% each time.

Outcome after 100 bets:

Compare to fixed 10% sizing:

Compare to fixed 50% sizing (overbetting):

Kelly's 20% delivers 30x more growth than half-Kelly and 4,000x more than double-Kelly. This isn't theoretical—it compounds in reality.

Applying Kelly to Real Tariff Trades

Let's work through actual examples using Kelly sizing.

Example 1: Section 301 Exclusion Renewal Trade

Setup (June 2024):

Kelly Calculation:

f* = (0.73 × $1.00 - $0.58) / $1.00
f* = $0.15 / $1.00
f* = 0.15 (15% of bankroll)

Trade:

Outcome: Bucket won (ETR settled 19.8%). Profit $3,150. Kelly sizing captured full edge while limiting downside risk.

Example 2: China-Mexico Spread Trade

Setup (January 2025):

Structure: Buy Mexico 2-5% bucket ($0.62), sell China 20-25% bucket ($0.54)

Kelly Calculation:

p = 0.65 (your probability estimate)
V = $1.00 (max payout)
C = $0.08 (cost)

f* = (0.65 × $1.00 - $0.08) / $1.00
f* = $0.57 / $1.00
f* = 0.57

Wait—57% of bankroll? This seems aggressive because the payoff ratio is extremely favorable (12.5:1 on your capital). Kelly says go big when you have high edge and high payoff.

Reality Check: Use fractional Kelly (discussed below). At quarter-Kelly: 0.57 / 4 = 14.25% of bankroll.

Example 3: Multi-Bucket Portfolio

Kelly also works for diversified positions across multiple buckets.

Setup:

Kelly for Each Bucket:

20-25% bucket:

f* = (0.61 × $1.00 - $0.52) / $1.00 = 0.09 (9%)

25-30% bucket:

f* = (0.22 × $1.00 - $0.35) / $1.00 = -0.13 (NEGATIVE - don't buy!)

15-20% bucket:

f* = (0.12 × $1.00 - $0.08) / $1.00 = 0.04 (4%)

Execution: Allocate 9% to 20-25%, 4% to 15-20%, zero to others. Total exposure: 13% of bankroll across two positions.

Why Fractional Kelly is Better for Real Trading

Full Kelly maximizes growth rate but creates stomach-churning volatility. A single bad run (even when you have edge) can draw down 30-50% of bankroll.

The Volatility Problem

Full Kelly simulation (100 trades, 60% win rate, 2:1 payoff):

Half-Kelly simulation (same 100 trades):

Quarter-Kelly simulation:

The trade-off: Full Kelly delivers 30x more growth than half-Kelly, but half-Kelly delivers 8x lower drawdown. For most traders, fractional Kelly between 1/4 and 1/2 optimizes risk-adjusted returns including psychological ability to stick with the strategy.

Recommended Fractional Sizing

Quarter-Kelly (1/4 Kelly): For most prediction market traders

Half-Kelly (1/2 Kelly): For experienced traders with high conviction

Full Kelly: Only for professional bankroll managers

More than Full Kelly: NEVER. Overbetting Kelly actually decreases long-run growth and increases ruin risk. If you think your edge is so good that you should bet 2x Kelly, you're probably overestimating your edge.

Estimating Your True Edge

Kelly is only as good as your probability estimates. Garbage in, garbage out.

How to Estimate Winning Probability

Method 1: Historical Base Rates

Method 2: Model-Based

Method 3: Market-Implied + Adjustment

Method 4: Ensemble

Calibration Testing

Track your past estimates vs outcomes to measure accuracy:

Calibration table: | Your Estimate | Actual Win Rate | Sample Size | |---------------|-----------------|-------------| | 50-60% | 52% | 18 trades | | 60-70% | 67% | 31 trades | | 70-80% | 71% | 14 trades |

If your "70% confident" trades only win 55% of the time, you're overconfident. Scale down estimates or use more conservative fractional Kelly.

Common Kelly Mistakes in Tariff Markets

Mistake 1: Ignoring Correlation

Kelly assumes independent bets. But if you have 5 positions on different China ETR buckets and they're all correlated with the same underlying policy variable, you're effectively making one big bet with 5x leverage.

Fix: Treat correlated positions as single bet for Kelly purposes. If you have 20-25% bucket and 25-30% bucket (both exposed to same USTR decision), calculate combined Kelly across both.

Mistake 2: Not Updating Estimates

Your initial 65% probability estimate was based on data available in January. It's now March, and new information emerged (USTR leaked memo, Congressional hearing testimony). If you don't update, your Kelly sizing is stale.

Fix: Recalculate Kelly weekly or after major news. If probability drops from 65% to 52%, reduce position size accordingly.

Mistake 3: Confusing Market Price with Fair Value

Market price ($0.54) doesn't tell you the TRUE probability—it tells you what the crowd thinks. Your edge comes from knowing better than the crowd.

Bad Kelly:

f* = (0.54 × $1 - $0.54) / $1 = 0 (no bet)

This assumes market price equals true probability, which means no edge.

Good Kelly:

Your probability: 68%
Market price: $0.54
f* = (0.68 × $1 - $0.54) / $1 = 0.14 (14% bet)

Mistake 4: Over-Leveraging on "Sure Things"

When you're 95% confident a trade will win, Kelly might say bet 90% of bankroll. Don't.

Reasons:

Fix: Cap Kelly at 25% of bankroll even when formula suggests more. Better to underbet "sure things" than overbet and get ruined.

Kelly for Different Tariff Trade Structures

Binary Trades (Will tariff be implemented?)

Use standard Kelly formula. Simple and clean.

Bucketed Scalars (Which ETR bucket?)

More complex because you're choosing between 4-6 buckets. Options:

  1. Calculate Kelly for each bucket independently, allocate to those with positive Kelly
  2. Construct portfolio of buckets that maximizes Kelly growth rate
  3. Use "most likely" bucket only (simplifies but leaves edge on table)

Recommendation: Option 1 is practical. Buy any bucket where f* > 0, skip others.

Calendar Spreads

Kelly applies to net payoff distribution, not individual legs.

Example: Long March $0.62, short June $0.54, net cost $0.08

Calculate expected value of all scenarios, use that for Kelly.

Cross-Country Spreads

Similar to calendar spreads—calculate distribution of net outcomes, apply Kelly to the spread as single bet.

Practical Kelly Sizing Worksheet

Step 1: Define the Trade

Step 2: Estimate Probability

Step 3: Calculate Full Kelly

f* = (0.61 × $1.00 - $0.52) / $1.00
f* = 0.09 (9% of bankroll)

Step 4: Apply Fractional Kelly

Step 5: Execute

Step 6: Track and Update

When Kelly Doesn't Apply

Kelly assumes:

  1. You know the true probabilities (or can estimate them accurately)
  2. You can make the same bet many times (law of large numbers)
  3. Payoffs are binary or knowable in advance
  4. You have unlimited ability to take the other side (no liquidity constraints)

Violations in tariff markets:

Liquidity: If you want to bet $50K but market only has $5K of liquidity, Kelly doesn't help. You're constrained by market size.

One-time bets: If this is your only tariff trade ever, Kelly's long-run growth rate optimization doesn't apply. Use expected value instead.

Unknown unknowns: Black swan executive orders, surprise trade deals, pandemic supply shocks—these aren't in your probability model. Kelly assumes you've captured all risks.

Correlated risks: If global trade war escalates and ALL your tariff positions lose simultaneously, Kelly's independence assumption breaks.

Fix: Use extra-fractional Kelly (1/6 or 1/8) when these violations are severe. Build in margin of error.

Conclusion: Kelly as Decision Framework

The Kelly Criterion isn't just a formula—it's a mindset. It forces you to:

  1. Quantify your edge: Can't apply Kelly without estimating probabilities, which means doing the analysis.
  2. Scale risk to opportunity: Bigger edge = bigger bet. Smaller edge = smaller bet. Eliminates emotional sizing.
  3. Avoid ruin: Kelly mathematically eliminates total loss risk (assuming positive edge exists).
  4. Calibrate over time: Tracking Kelly estimates vs outcomes reveals if you're overconfident or underconfident.

For tariff prediction markets where genuine information edges exist (USTR docket analysis, port trade flow tracking, Congressional activity monitoring), Kelly sizing converts that edge into optimal long-term growth.

Start with quarter-Kelly. Track your calibration. Adjust fraction based on drawdown tolerance. Avoid overbetting. And remember: position sizing matters more than win rate—bet too much on your winners and you'll blow up; bet too little and you'll never compound.

The traders who survive 10 years in prediction markets aren't the ones with the highest win rates. They're the ones who sized their positions correctly.

Sources

Risk Disclosure

The Kelly Criterion assumes accurate probability estimation, which is extremely difficult in prediction markets. Overestimating probabilities while using Kelly sizing can lead to catastrophic losses. This analysis is for educational purposes only and does not constitute investment advice. Prediction markets involve substantial risk. Always use fractional Kelly and never bet more than you can afford to lose.

Ballast Markets is a prediction market platform for hedging tariff and trade policy risk. Learn more at ballastmarkets.com.