Portfolio Construction: Building an Edge-Based System
Your edge isn't in ONE great trade—it's in how you allocate capital across HUNDREDS of trades.
Professional traders don't just find good setups. They build portfolios that maximize expected returns while controlling risk. This lesson teaches you how to construct a portfolio like an institutional fund manager.
🎯 What You'll Learn
By the end of this lesson, you'll be able to:
- Calculate Kelly Criterion position size for any trading strategy
- Understand why full Kelly is dangerous (and what fractional Kelly to use)
- Build multi-strategy portfolios with proper correlation adjustments
- Implement dynamic position sizing based on equity curve performance
- Use Signal Pilot tools to validate edge and optimize allocations
⚡ Quick Wins for Tomorrow (Click to expand)
Start with these 3 actions:
- Calculate Kelly % for your best strategy. Need 3 numbers: (1) Win rate (wins ÷ total), (2) Avg winner, (3) Avg loser. Formula: Kelly % = (WR × (AvgW ÷ AvgL) - (1-WR)) ÷ (AvgW ÷ AvgL). Example: 58.3% WR, $360 avg win, $200 avg loss = 35.1% full Kelly. Use 1/4 Kelly = 8.8% per trade. Ed Thorp ran Princeton/Newport Partners on fractional Kelly at roughly 20% a year for two decades.
- Implement equity curve position sizing. Calculate 20-day MA of equity. Above MA = increase size 25% (hot streak). Below MA = decrease 50% (cold streak). Prevents revenge trading, maximizes winners. It is the oldest discipline rule on a professional desk: press when the book is working, cut when it is not.
- Run Kelly stress test on worst losing streak. Find worst consecutive losses (say 7). Calculate: 2% risk = -14% DD. Full Kelly 20% = -79% DD (account destroyed). Quarter Kelly 5% = -30% DD (recoverable). Full Kelly WILL destroy you. Use quarter Kelly.
🧮 Calculate Your Kelly Percentage
Stop guessing position size. Use the Kelly Criterion to mathematically optimize your risk based on your proven edge. Input your win rate and average R:R to get full Kelly, half Kelly, and quarter Kelly percentages.
Use Kelly Criterion Calculator →💰 How Kelly Criterion Saved a $500K Account
Prop trader on a $500K book: 58% WR, $450 avg win, $300 avg loss. Old (2% fixed): +$140K/year (+28%), Sharpe 1.2, 15% DD. The calculation: b = 450/300 = 1.5, so full Kelly = (0.58 × 1.5 − 0.42) / 1.5 = 30.0% — a number he was never going to trade. He moved to one-tenth of it: 3% per trade. New (3%): +$224K/year (+45%), Sharpe still 1.2, 22% DD. Sizing does not improve your risk-adjusted return — return and drawdown both scaled by roughly 1.5×. What Kelly bought him was the confidence that 3% was nowhere near the ceiling, so he could stop leaving two-thirds of a proven edge unused.
📉 CASE STUDY: Lisa's $114,000 Full Kelly Disaster
Setup: Lisa Chang (composite example), options trader with 68% WR credit spreads — the shape credit spreads actually have: small wins, big losses (avg win $250, avg loss $400, 0.63:1 R:R). Expectancy +$42 per trade, a real edge. Full Kelly = (0.68 × 0.625 − 0.32) / 0.625 = 16.8% per trade. $185K account, Apr-Jun 2024.
The disaster: Weeks 1-2: +$7.5K profit using full Kelly, account at $192.5K. Week 3: a 6-trade losing streak (0.32^6 = 0.1% probability). Each loss takes 16.8%, compounding: 0.832^6 = 0.332, a -66.8% drawdown. Account dropped $192.5K → $64K in ONE WEEK. She traded small for the rest of the quarter and finished June at $71K: -$114K, -62% in 3 months, on a strategy whose edge was never in doubt.
Recovery: Switched to 1/4 Kelly (4.2% per trade). Results: $71K → $111K (+56%) in 6 months, max DD -12% (vs -67%). Same edge, survivable variance. The identical six-loss streak at 4.2% would have taken her from $192.5K to $149K instead of $64K — the fraction was worth $85K.
Lesson: Full Kelly is the theoretical optimum and it carries catastrophic variance. Quarter Kelly keeps about 44% of the theoretical growth rate with 12% drawdowns instead of 67% — and 44% of a growth rate you survive beats 100% of one you do not. Calculate full Kelly, then USE 25-50% of that number.
Case Study Quiz: Lisa had a profitable strategy (68% win rate, 0.63:1 R:R, +$42 expectancy per trade) and calculated her optimal Kelly Criterion as 16.8% per trade. She used FULL Kelly and lost $114K (-62%) in 3 months despite having positive edge. What was her fatal mistake?
The Single-Strategy Trap
Most retail traders optimize for ONE perfect setup. They spend months backtesting a breakout strategy, achieve 55% win rate with 2:1 R:R, and think they've "made it."
The problem: Single-strategy traders are one market regime change away from blowing up.
📉 Real Example: The Breakout Trader Whose Edge Vanished in One Quarter
Setup: Trader had a proven momentum breakout system. 2022 results: 58% win rate, $580 avg win, $320 avg loss — expectancy +$202 per trade. Over 233 trades that made $47K on a $150K account (+31%), finishing the year at $197K.
January-April 2023: Fed pivots hawkish. The market shifts from trending (breakouts work) to choppy and range-bound (breakouts fail to follow through).
Result: Win rate fell from 58% to 38% and the average winner shrank from $580 to $310, because breakouts stopped extending. Expectancy flipped from +$202 to -$80 per trade. He traded through it anyway and doubled his size in March to make it back: 190 trades, -$20K (-10% of the $197K he had built) — and, more expensively, four months of the year earning nothing, because he had ZERO mean-reversion strategies to capitalize on the new regime.
Lesson: One strategy = one point of failure. The regime change did not blow up his account — it deleted his edge, which for a one-strategy trader is the same thing. Institutions run 5-15 uncorrelated strategies so that a dead regime costs them one sleeve, not the whole year.
The Three Pillars of Portfolio Construction
Professional portfolio construction rests on three pillars:
| Pillar | Retail Approach | Institutional Approach |
|---|---|---|
| 1. Diversification | One "perfect" strategy | 5-10 strategies across regimes (trend, mean-reversion, volatility, correlation) |
| 2. Position Sizing | Fixed 1-2% per trade (ignores edge) | Kelly Criterion adjusted for edge strength (5-15% for proven edges) |
| 3. Correlation Management | Ignored (all strategies correlated) | Reduce allocation to correlated strategies (avoid doubling down on same bet) |
Edge-Based Allocation vs Equal Weighting
Equal weighting: Allocate same % to each strategy (e.g., 3 strategies = 33% each)
Edge-based allocation (Kelly): Allocate MORE to strategies with stronger edge, LESS to weaker edges
Example: Why Equal Weighting Leaves Money on the Table
3 strategies: A (65% WR, 1.2:1 → full Kelly 35.8%), B (52% WR, 1.5:1 → 20.0%), C (48% WR, 1.3:1 → 8.0%). Equal weight: 33% of the risk budget each — you put the same money behind C, whose edge is a quarter of A's. Kelly weight: allocate in proportion to edge — 56% / 31% / 13%. Growth is roughly proportional to the risk-weighted average edge: equal weighting gives (35.8 + 20.0 + 8.0) / 3 = 21.3, edge weighting gives 0.56 × 35.8 + 0.31 × 20.0 + 0.13 × 8.0 = 27.3. That is about 28% more expected growth from the same total risk, because every dollar moved from C to A is a dollar earning four and a half times the edge. Allocate based on edge, not equality.
Building for Regime Resilience
Markets cycle through four primary regimes. Your portfolio should have strategies for each:
| Regime | Characteristics | Strategy Types That Work |
|---|---|---|
| Trending Bull | Higher highs, low volatility, QE environment | Momentum breakouts, pullback buying, BTFD strategies |
| Trending Bear | Lower lows, rising volatility, QT environment | Breakdown shorts, rally fades, put spreads |
| Range-Bound | Choppy, oscillating between support/resistance | Mean reversion, support/resistance trades, theta decay (options selling) |
| High Volatility | VIX > 25, whipsaw moves, uncertainty | Volatility selling (after spikes), wide stop strategies, reduced size |
💡 Pro Tip: The "2-2-1" Portfolio Structure
Institutions often use a "2-2-1" diversification model:
- 2 Trend strategies (long bias and short bias) — capitalize on directional moves
- 2 Mean-reversion strategies (support bounces and resistance fades) — profit from ranges
- 1 Volatility strategy (VIX spike trading or premium selling) — hedge against chaos
This ensures you have exposure to multiple regimes simultaneously. When trends die, mean-reversion activates. When volatility explodes, volatility strategy hedges the others.
What Is Kelly Criterion?
The Kelly Criterion is a mathematical formula that answers one question: "Given my edge (win rate and reward/risk ratio), what percentage of my account should I risk per trade to maximize long-term compound growth?"
Developed by John Kelly Jr. at Bell Labs in 1956, it's been used by:
- Ed Thorp: Beat blackjack with card counting, then ran Princeton/Newport Partners on fractional Kelly at roughly 20% a year for two decades — 227 profitable months out of 230
- Warren Buffett and Charlie Munger: Both have described sizing by how strong the edge is rather than by fixed weights — Kelly logic without the formula
- Renaissance Technologies: Jim Simons' Medallion fund compounded at roughly 39% a year net of fees for three decades on exactly this shape — many small, weakly correlated edges, each sized for growth rather than for comfort
The Kelly Formula (Simplified)
Kelly % = (Win Rate × Avg Win - Loss Rate × Avg Loss) / Avg Win
Or, in mathematical notation:
f* = (p × b - q) / b
Where:
- f* = Optimal fraction of capital to risk (Kelly %)
- p = Win rate (probability of winning)
- q = Loss rate (1 - p)
- b = Win/Loss ratio (Avg Win / Avg Loss)
Step-by-Step Kelly Calculation
Example: 60% win rate, $500 avg win, $300 avg loss. Variables: p = 0.60, q = 0.40, b = $500/$300 = 1.67. Formula: f* = (0.60 × 1.67 - 0.40) / 1.67 = 0.60 / 1.67 = 0.359 = 35.9% full Kelly. Use Quarter Kelly: 35.9% / 4 = 9% per trade.
✅ What This Means in Practice
With a $100,000 account and this 60% win rate system:
- Full Kelly (35.9%): Risk $35,900 per trade — suicidal (one bad week destroys account)
- Quarter Kelly (9%): Risk $9,000 per trade — optimal balance of growth and survival
- Typical 1-2% sizing: Risk $1,000-$2,000 per trade — safe but leaves massive edge on the table
Key insight: If you have a PROVEN 60% win rate with 1.67:1 R:R, risking only 1-2% leaves most of your edge unused. The growth-optimal number is 35.9%; the number you would actually trade is a quarter of it. The next section shows exactly why you take the quarter and not the whole.
Why Full Kelly Is Dangerous (The Variance Problem)
Full Kelly optimizes for maximum geometric growth but ignores psychological survival.
The math problem: Even with a 60% win rate, you WILL hit losing streaks. Probability of 5 consecutive losses:
0.40^5 = 0.01024 = 1.024%
Translation: With 100 trades per year, you'll hit a 5-trade losing streak roughly once per year.
| Kelly Fraction | Risk Per Trade | Loss After 5-Trade Streak | Recovery Needed |
|---|---|---|---|
| Full Kelly (35.9%) | $35,900 | -89.2% (account at $10,800) | +824% to break even (impossible) |
| Half Kelly (18%) | $18,000 | -62.9% (account at $37,100) | +170% to break even (very hard) |
| Quarter Kelly (9%) | $9,000 | -37.6% (account at $62,400) | +60% to break even (achievable) |
| Fixed 2% | $2,000 | -9.6% (account at $90,400) | +11% to break even (easy, but slow growth) |
Conclusion: Growth at a fraction f of Kelly is (2f − f²) of the maximum. Half Kelly keeps 75% of the optimal growth rate at half the volatility; quarter Kelly keeps 44% at a quarter of the volatility. Full Kelly keeps 100% of the growth rate and hands you the drawdowns in the table above. Giving up a quarter of your growth to halve your swings is the trade almost everyone should take.
Why Ed Thorp Used Fractional Kelly
Thorp — the mathematician who beat blackjack and then ran Princeton/Newport at roughly 20% a year for two decades — argued for fractional Kelly on arithmetic, not nerves. Growth at a fraction f of Kelly is (2f − f²) of the maximum, while volatility scales linearly with f. At half Kelly you keep 2(0.5) − 0.25 = 75% of the growth rate for half the swings. At quarter Kelly you keep 2(0.25) − 0.0625 = 44% for a quarter of the swings. Paying a quarter of your growth rate to halve your drawdowns is the best trade on the menu, and paying 56% of it to quarter them is the one most people should still take — the drawdowns you actually live through decide whether you are around to compound at all. Rule: Use 25-50% of Kelly for real trading.
Multi-Strategy Portfolio Allocation
Real portfolios don't have just one strategy. You might run momentum breakouts, mean-reversion trades, and options income simultaneously.
The question: How do you allocate capital across multiple strategies with different edges?
Step 1: Calculate Kelly % for Each Strategy Independently
Example portfolio with 3 strategies:
| Strategy | Win Rate | Avg Win | Avg Loss | Win/Loss Ratio | Full Kelly % | Quarter Kelly % |
|---|---|---|---|---|---|---|
| A: Breakouts | 45% | $600 | $200 | 3.0 | 26.7% | 6.7% |
| B: Mean Rev | 65% | $300 | $250 | 1.2 | 35.8% | 8.95% |
| C: Volatility | 55% | $450 | $350 | 1.29 | 20.0% | 5.0% |
Total allocation (if strategies were independent): 6.7% + 8.95% + 5.0% = 20.65%
Step 2: Adjust for Correlation
The problem: If your strategies are correlated (win/lose together), you're effectively making the same bet multiple times.
Correlation coefficient ranges from -1 to +1:
- +1.0: Perfect correlation (always move together) — same strategy, don't double allocation
- +0.6 to +0.8: High correlation (often move together) — reduce allocation by 30-50%
- +0.2 to +0.4: Low correlation (somewhat independent) — reduce allocation by 10-20%
- 0.0: No correlation (truly independent) — no adjustment needed
- -0.5 to -1.0: Negative correlation (move opposite) — can increase total allocation (hedged)
How to Calculate Strategy Correlation (Excel/Python)
Excel: Export daily P&L to columns (Date, Strategy A, Strategy B). Use =CORREL(B:B, C:C) for correlation coefficient.
Python: df[['Strategy_A', 'Strategy_B']].corr() gives correlation matrix.
Interpretation: 0.15 = nearly independent ✅. 0.60 = high overlap ⚠️ (reduce allocation). 0.05 = independent ✅.
Step 3: Apply Correlation Adjustment
Continuing our 3-strategy example, assume these correlations:
- Strategy A & B: 0.15 (low correlation)
- Strategy B & C: 0.60 (high correlation) ⚠️
- Strategy A & C: 0.05 (independent)
Adjustment rule: Reduce allocation to correlated pairs by 30% (for 0.60 correlation)
| Strategy | Quarter Kelly % | Correlation Issue | Adjusted Allocation |
|---|---|---|---|
| A: Breakouts | 6.7% | None (independent) | 6.7% (no change) |
| B: Mean Rev | 8.95% | 0.60 with C | 8.95% × 0.70 = 6.27% |
| C: Volatility | 5.0% | 0.60 with B | 5.0% × 0.70 = 3.5% |
New total allocation: 6.7% + 6.27% + 3.5% = 16.47%
Result: Reduced from 20.65% to 16.47% to account for B-C correlation. This prevents over-concentration when those two strategies lose together.
⚠️ The "Diversification Illusion" Trap
Many traders think they're diversified because they trade "stocks, crypto, and forex" — but all three are risk-on assets that crash together during Fed hawkishness.
Example: November 2021 - January 2022
- The Fed signals tapering, then accelerates it — the market starts pricing QT
- From the November 2021 highs to the late-January 2022 lows: SPY -10.6%, ARKK -51%, BTC -52%
- A trader long US equities, growth names and crypto lost on every sleeve in the same eleven weeks — three "different" markets, one Fed trade
True diversification = strategies that profit in DIFFERENT market conditions:
- Long breakouts (profit in QE bull markets)
- Short fades (profit in QT bear markets)
- Mean reversion (profit in choppy ranges)
- Volatility selling (profit when VIX spikes then collapses)
Step 4: Set Portfolio Heat Limits
Portfolio heat = Total risk across all open positions simultaneously
Institutional standard: Never exceed 10-15% total portfolio heat
Example:
- Strategy A: 6.7% allocation, currently have 1 trade open → 6.7% heat
- Strategy B: 6.27% allocation, currently have 2 trades open → 12.54% heat
- Strategy C: 3.5% allocation, currently have 1 trade open → 3.5% heat
Total portfolio heat: 6.7% + 12.54% + 3.5% = 22.74% ⚠️ TOO HIGH!
Action required: Don't take new trades until portfolio heat drops below 15%. Close or scale out of positions to reduce exposure.
💡 Pro Tip: The "Portfolio Heat Dashboard"
Create a simple spreadsheet that calculates real-time portfolio heat:
| Strategy | Kelly % | Open Positions | Current Heat |
|---|---|---|---|
| Breakouts | 6.7% | 1 | 6.7% |
| Mean Rev | 6.27% | 0 | 0% |
| Volatility | 3.5% | 1 | 3.5% |
| Total Portfolio Heat | 10.2% ✅ | ||
Rule: Before entering any new trade, check this dashboard. If adding the trade would push heat above 15%, wait for an existing position to close first.
Part 4: Dynamic Position Sizing
Fixed-Fraction vs Dynamic Kelly
Fixed-Fraction: Always risk same % (e.g., 1% per trade)
Dynamic Kelly: Adjust size based on recent performance
Equity Curve Trading
Concept: Increase size when winning, decrease when losing
Method:
- Calculate 20-day moving average of equity curve
- If equity > MA: Increase position size by 25% (edge working)
- If equity < MA: Decrease position size by 50% (edge broken or regime shift)
Benefit: Automatically scales risk to match performance (compound winners, limit losers)
Example: Equity Curve Position Adjustment
Baseline: $100K account, 1% risk = $1,000/trade. Hot streak (equity > MA): $108K account, 1.25% risk = $1,350/trade. Cold streak (equity < MA): $97K account, 0.5% risk = $485/trade. Risk more when winning, less when losing.
Part 5: Using Signal Pilot for Portfolio Construction
Janus Atlas: Multi-Strategy Overlay
Feature: Visualize all active strategies on same chart (identify correlation)
Use case: If all your setups trigger on same day → high correlation → reduce total size
Pentarch Pilot Line: Edge Validation
Feature: Compare your entries vs institutional flow
Signal: If Pilot Line confirms your setup (institutional flow aligned) → increase size to Kelly %
Warning: If Pilot Line contradicts setup (institutions selling, you buying) → reduce size to 0.5× Kelly
Harmonic Oscillator: Regime Detection for Sizing
Feature: Identify trending vs mean-reverting regimes
Application: Increase breakout strategy allocation in trending regime, increase mean-reversion allocation in ranging regime
💡 Pro Tip: The "Correlation Audit" Every Quarter
Most traders blow up from hidden correlation. March 2023, SVB week: a trader ran 4 "uncorrelated" strategies — long regional banks, long energy, short credit spreads, short vol. Historical pairwise correlation 0.2-0.3. In the crisis week every one of them was the same trade: KRE fell 28%, energy sold off with it, credit spreads widened, and VIX ran from 19 to 26. All four lost together, -22% across the book in a week. Audit: if the Fed moves 0.5%, which strategies lose? If VIX spikes to 40, how many stops get hit? If more than 60% of your book fails the same scenario, you are not diversified, you are concentrated with extra steps.
🎯 Practice Exercise: Build Your Kelly Portfolio
Scenario: Your Three Trading Strategies
You have $100,000 capital and three proven strategies. Calculate optimal Kelly allocation for each:
| Strategy | Win Rate | Avg Win | Avg Loss | Trades/Month |
|---|---|---|---|---|
| Strategy A: Momentum Breakouts | 45% | $600 | $200 | 8 |
| Strategy B: Mean Reversion | 65% | $300 | $250 | 12 |
| Strategy C: Options Income | 70% | $200 | $400 | 15 |
Additional Info:
- Correlation between Strategy A & B: 0.15 (nearly independent)
- Correlation between Strategy B & C: 0.6 (moderately correlated)
- Correlation between Strategy A & C: 0.05 (independent)
Your Tasks:
Task 1: Calculate full Kelly % for each strategy
Formula: Kelly % = (Win Rate × (Avg Win / Avg Loss) - (1 - Win Rate)) / (Avg Win / Avg Loss)
Task 2: Which fractional Kelly should you use? (Full, Half, or Quarter?)
Task 3: Adjust allocations for correlation. Which strategies need reduced sizing?
Task 4: Calculate dollar amount to risk per trade for each strategy
📋 Solution (Try First!)
Click to Reveal Solution
Task 1: Full Kelly: A = 26.7% (45% WR, 3.0 b), B = 35.8% (65% WR, 1.2 b), C = 10.0% (70% WR, 0.5 b).
Task 2: Use quarter Kelly — about 44% of the theoretical growth rate for 10-15% drawdowns instead of the 40-70% full Kelly produces. Result: A = 6.7%, B = 9.0%, C = 2.5%.
Task 3: B and C have 0.6 correlation, so both take the 30% haircut from Part 3: A = 6.7% (unchanged), B = 9.0% × 0.70 = 6.3%, C = 2.5% × 0.70 = 1.75%. Total = 14.75%.
Task 4: $100K account: A = $6,700/trade, B = $6,300/trade, C = $1,750/trade — $14,750 of heat if all three are open at once, inside a 15% cap. If the edges hold, expect roughly 15-25% CAGR with 12-18% peak-to-trough drawdowns.
📝 Knowledge Check
Test your understanding of Kelly Criterion and portfolio construction:
Your trading system shows: 60% win rate, $400 average win, $200 average loss. What's the correct Kelly % and what should you actually risk per trade?
You have 3 trading strategies with these full Kelly %: Strategy A = 20%, Strategy B = 25%, Strategy C = 30%. Average correlation between them is 0.80. What's your total portfolio allocation?
Your account equity curve has dropped below its 20-day moving average for the first time in 3 months. Your standard position size is 2% per trade. What should you do?
Portfolio Construction Workflow:
- Backtest each strategy: Calculate win rate, win/loss ratio
- Calculate Kelly % for each strategy (use quarter or half Kelly)
- Identify correlation between strategies (do they win/lose together?)
- Reduce allocation to highly correlated strategies (correlation >0.6)
- Cap total portfolio heat at 10-15% — allocation is not the same thing as heat
- Implement equity curve tracking (20-day MA)
- Adjust sizing based on equity vs MA (increase when above, decrease when below)
Pre-Trade Kelly Check:
- Identify which strategy this setup belongs to
- Check remaining allocation for that strategy (have you hit Kelly limit?)
- Check portfolio heat: Total risk across all open positions under your 10-15% cap?
- Use Signal Pilot Pentarch Pilot Line to confirm edge (institutional flow aligned?)
- If edge confirmed, use full (quarter/half) Kelly. If not, use 0.5× Kelly or skip
Key Takeaways
- Kelly Criterion mathematically optimizes sizing based on edge (win rate × win/loss ratio)
- Never use full Kelly (too volatile) — use 0.25-0.5× Kelly
- Correlation matters: Reduce allocation to correlated strategies (they're the same bet)
- Dynamic sizing: Increase when equity > MA, decrease when < MA
- Portfolio heat cap: Total risk across all open positions should never exceed 10-15%
- Overconfidence kills: 60% WR + 2:1 R:R = 40% full Kelly. A 7-loss streak (0.4^7 ≈ 0.16% probability, so about once every 600 trades) leaves you down 97% at full Kelly and 52% at quarter Kelly. Kelly names the ceiling, not the target — most professionals run well under quarter Kelly and cap total heat besides.
Kelly criterion provides the math, fractional Kelly provides survival. Build portfolios that grow steadily without ruin risk.
Related Lessons
Advanced Risk Management
Foundation for Kelly-based portfolio construction.
Read Lesson →Quantitative Strategy Design
Advanced quantitative methods for portfolio optimization.
Read Lesson →⏭️ Coming Up Next
Lesson #48: Institutional Order Flow — Enter the Advanced track and learn how institutions execute large orders without moving markets.
Downloads
Educational only. Trading involves substantial risk of loss. Past performance does not guarantee future results.
💬 Discussion (0 comments)
Loading comments...
Ready to Trade with Signal Pilot?
Apply your trading education with professional indicators and real-time market analysis tools.
Back to Signal Pilot →