Chapter 17 Key Takeaways: Portfolio Construction and Risk Management

The Big Ideas

1. Survival Is the Prerequisite for Success

The single most important insight in this chapter: no amount of edge matters if a drawdown eliminates your ability to trade. Portfolio construction and risk management exist to ensure that you stay in the game long enough for your skill to compound. A trader with modest edge and excellent risk management will outperform a trader with large edge and no risk management over any sufficiently long time horizon.

2. Portfolio Thinking Transforms Prediction Market Trading

Single-trade thinking asks "How much should I bet on this market?" Portfolio thinking asks "How should I allocate my total bankroll across all available opportunities, given their edges, correlations, and my risk tolerance?" This shift unlocks diversification benefits, enables systematic risk measurement, and produces smoother returns through the law of large numbers.

3. Correlation Is the Hidden Risk in Prediction Markets

Two positions in the same political cycle, two economic indicators driven by the same macro environment, or two sports bets in the same tournament may appear independent but share common drivers. The Pearson correlation between binary events has narrower bounds than for continuous variables, making it harder to achieve extreme correlation --- but even moderate correlations dramatically increase portfolio risk during stress events. Correlations are not constant: they spike during crises, turning "diversified" portfolios into concentrated bets.

4. Portfolio Kelly Accounts for What Single Kelly Cannot

Applying the single-bet Kelly Criterion independently to each opportunity ignores three realities: the budget constraint (you cannot wager more than 100% of your bankroll), correlation effects (correlated bets compound risk), and interaction effects (optimal sizing for Bet A depends on your allocation to Bet B). Portfolio Kelly jointly optimizes all allocations using Monte Carlo methods and numerical optimization.

5. Fractional Kelly Is Not Optional in Portfolios

Estimation errors in probabilities and correlations compound across positions. If one bet's probability is 5% off, the effect is modest. If twenty bets' probabilities are each 5% off, the portfolio-level impact can be severe. Using 25-50% of full Kelly provides a crucial buffer against model misspecification while sacrificing relatively little expected growth.

6. Diversification Is the Only Free Lunch

Combining uncorrelated positive-edge bets reduces portfolio variance without reducing expected return. A portfolio of 50 uncorrelated bets with 5-cent average edge has the same expected return as concentrating in 5 bets, but with dramatically lower variance. The diversification ratio quantifies this benefit: values above 1.5 indicate strong diversification. Diversify across event types, time horizons, platforms, and analytical strategies.

7. Risk Metrics Must Be Adapted for Binary Outcomes

Standard risk metrics (VaR, Sharpe Ratio, drawdown) were designed for continuous return distributions. Binary prediction markets produce discrete outcomes, requiring adapted computation methods --- primarily Monte Carlo simulation. Expected Shortfall (CVaR) is more informative than VaR because it captures the severity of tail losses, not just their threshold.

8. Drawdown Management Requires Pre-Commitment

Drawdowns are inevitable for every trader, regardless of skill. The question is not whether they will occur but how deep, how long, and whether you survive them. Pre-committed rules --- graduated position reductions at 10%, 15%, 25%, and 40% drawdown thresholds --- prevent both panic selling and denial. These rules must be written before you need them, when you are thinking rationally.

9. Bankroll Structure Ensures Survival

Never deploy all available capital. Structure your money into tiers: Trading Capital (40-60%, actively deployed), Reserve Capital (20-30%, available to replenish losses), and Emergency Fund (10-20%, untouchable). The reserve is not idle money --- it is insurance that keeps you solvent through any single adverse event.

Formulas to Remember

Formula Use Case
$\rho_{XY} = \frac{P(X=1,Y=1) - p_X p_Y}{\sqrt{p_X(1-p_X) \cdot p_Y(1-p_Y)}}$ Pearson correlation between binary events
$G(\mathbf{f}) = \sum_s P(s) \log(1 + \sum_i f_i R_i(s))$ Portfolio Kelly objective function
$\text{Var}(R_p) \to \bar{\rho} \cdot p(1-p) \cdot ((1-p)/p)^2$ Irreducible variance with correlation $\bar{\rho}$
$\text{VaR}_\alpha = -\text{Quantile}_{1-\alpha}(R_p)$ Value at Risk
$\text{ES}_\alpha = -E[R_p \mid R_p \leq -\text{VaR}_\alpha]$ Expected Shortfall
$\text{DR} = \frac{\sum w_i \sigma_i}{\sigma_p}$ Diversification ratio
$\text{MDD} = \max_t \frac{\max_{s \leq t} V_s - V_t}{\max_{s \leq t} V_s}$ Maximum drawdown
Recovery rounds $= \frac{\log(1/(1-d))}{\log(1+\mu)}$ Rounds to recover from drawdown $d$

Practical Checklist

Before deploying a prediction market portfolio, verify:

  • [ ] Correlations estimated: Every position pair has a correlation estimate based on structural analysis, conditional probabilities, or market-implied data
  • [ ] Portfolio Kelly computed: Allocations jointly optimized accounting for correlations and budget constraint
  • [ ] Fractional Kelly applied: Using 25-50% of full Kelly fractions
  • [ ] Individual position caps enforced: No single position exceeds 4-5% of bankroll
  • [ ] Correlated group caps enforced: No correlated group exceeds 10-20% of bankroll
  • [ ] Platform caps enforced: No single platform holds more than 25-40% of capital
  • [ ] Aggregate deployment capped: Total deployment does not exceed 50-80% of trading capital
  • [ ] Monte Carlo simulation run: Verified return distribution, VaR, CVaR, and probability of profit
  • [ ] Stress tests completed: Correlation spike, edge reduction, category wipeout, platform failure
  • [ ] Drawdown rules written: Specific thresholds and position scaling factors documented in advance
  • [ ] Bankroll tiers established: Trading, reserve, and emergency capital separated
  • [ ] Rebalancing triggers defined: Criteria for closing, adjusting, and opening positions documented

Common Mistakes to Avoid

  1. Sizing each bet independently with single Kelly. This ignores correlations and the budget constraint, potentially allocating well over 100% of your bankroll. Always use portfolio Kelly or apply position caps.

  2. Underestimating correlations. Traders routinely treat weakly related events as independent. Two Senate races in the same cycle are not independent. Two economic indicators in the same quarter are not independent. When in doubt, overestimate correlations.

  3. Using full Kelly. Full Kelly maximizes expected logarithmic growth but with enormous variance. In a portfolio with imperfect correlation and probability estimates, full Kelly is a recipe for ruin. Use half-Kelly or less.

  4. Concentrating in your area of expertise. If your edge is in political markets, the temptation is to allocate heavily to politics. But a portfolio concentrated in one category sacrifices the diversification benefit and creates catastrophic vulnerability to category-specific shocks.

  5. Ignoring platform risk. Capital on a prediction market platform is subject to the platform's solvency, regulatory compliance, and technical reliability. Spreading across platforms is essential even if it complicates operations.

  6. Treating correlations as fixed. Correlations estimated during calm periods dramatically understate correlations during crises. Stress test with correlation levels 3-5x your base estimates.

  7. Abandoning drawdown rules during a drawdown. The whole point of pre-committed rules is that they override emotional decision-making during stressful periods. If you override your rules when they activate, you might as well not have them.

  8. Chasing recovery by increasing size. After a drawdown, the temptation to "win it back quickly" by increasing bet sizes is the fastest path to ruin. Drawdown rules prescribe the opposite: reduce size and let compounding do the recovery work.

  9. Neglecting reserve capital. Deploying 100% of available capital feels productive but eliminates your margin of safety. Reserves are not idle; they are the insurance premium that keeps you in the game.

  10. Skipping stress tests. A Monte Carlo simulation under base assumptions is necessary but insufficient. If you have not tested what happens when correlations spike, edges disappear, or a platform fails, you are driving without a seatbelt.