Skip to main content
Common Risk-Management Mistakes

Why Traders Under-Size Their Highest-Conviction Ideas

Pomegra Learn

Why Does Conservative Position Sizing on Your Best Ideas Destroy Long-Term Returns?

You've identified a trade setup that your system has crushed for the past three years. In backtests spanning 500+ trades, it wins 68% of the time with an average win of $850 against an average loss of $600. The win-loss ratio is nearly 1.4:1. Your research shows the edge is clean, robust, and likely to persist. Yet when the setup appears in real trading, you size it at 1 contract—the same size as your weakest, lowest-confidence trades. You have the same risk exposure on a 55%-win setup ($1,000 avg win, $900 avg loss) as on your 68%-win edge. This is under-sizing your winners, and it's a hidden leak in compounding returns that often destroys long-term profitability more than actual losses do.

Quick definition: Under-sizing your best ideas means deploying capital that's proportional to risk rather than proportional to edge. You keep risk constant (e.g., $1,000 per trade) across all trades, ignoring that trades with higher win rates and larger win-loss ratios justify larger positions.

Key takeaways​

  • A 68% win-rate edge sized identically to a 55% win-rate edge leaves 30-40% of expected return on the table each year
  • Kelly Criterion formula (edge sizing formula used by casinos, hedge funds, and professional gamblers) recommends 3-5x larger position sizes on high-edge trades
  • Fixed-risk sizing (risk $1,000 per trade, always) is overly conservative for traders with measurable, asymmetric edges
  • The biggest returns come from "boring" statistical edges (55%-58% win rate, 1.8:1 risk-reward) sized properly, not from rare, spectacular trades
  • Ironically, proper sizing of your best ideas reduces risk (better return per unit of volatility, higher Sharpe ratio) while simultaneously increasing returns

The Return Leakage: Mathematics of Undersized Edges​

Assume you have two trade setups:

Setup A (Your Edge): 68% win rate, $850 avg win, $600 avg loss, 1.42:1 ratio

  • Expected profit per trade = (0.68 × $850) + (0.32 × -$600) = $578 + (-$192) = $386 per trade

Setup B (Weak Edge): 55% win rate, $1,000 avg win, $900 avg loss, 1.11:1 ratio

  • Expected profit per trade = (0.55 × $1,000) + (0.45 × -$900) = $550 + (-$405) = $145 per trade

Setup A has 2.66x the expected value of Setup B. Yet if you size both at 1 contract ($1,000 risk), you're treating them identically. Over 100 trades in each setup (200 total), your income is:

  • Setup A: 100 × $386 = $38,600
  • Setup B: 100 × $145 = $14,500
  • Total: $53,100

But what if you sized appropriately?

Setup A justifies 2.5x position size (per Kelly Criterion rough estimation). Setup B stays at 1x.

  • Setup A: 100 × ($386 × 2.5) = $96,500
  • Setup B: 100 × $145 = $14,500
  • Total: $111,000

You've left $57,900 on the table (52% profit upside) by under-sizing your edge.

Over 10 years (2,000 A-trades, 2,000 B-trades), the difference compounds:

  • Conservative (equal sizing): $53,100/year × 10 = $531,000 total profit
  • Proper sizing (kelly-adjusted): $111,000/year × 10 = $1,110,000 total profit

The proper-sizing strategy compounds from $100,000 → $1.2 million. The conservative strategy compounds from $100,000 → $631,000. Not sizing your edge is equivalent to leaving $570,000 of your future wealth on the table.

Kelly Criterion: The Formula Professional Gamblers Use​

The Kelly Criterion is the mathematically optimal formula for position sizing given an edge. It was developed by John Kelly at Bell Labs in 1956 and is used by:

  • Professional sports bettors (Las Vegas, DraftKings)
  • Hedge funds and CTAs
  • Casinos (they use it in reverse, to size house bets)
  • Professional poker players (tournament sizing)
  • Insurance companies (premium sizing)

The formula:

f* = (p × b - q) / b

Where:
f* = fraction of capital to bet
p = probability of win (0.68 for our Setup A)
q = probability of loss (0.32)
b = ratio of win to loss (850/600 = 1.417)

For Setup A:

f* = (0.68 × 1.417 - 0.32) / 1.417
f* = (0.963 - 0.32) / 1.417
f* = 0.643 / 1.417
f* = 0.454 = 45.4%

This means: bet 45.4% of your capital on each trade in Setup A. If you're managing a $100,000 account, each Setup A trade should risk $45,400.

Wait—that sounds crazy. Doesn't it? That's because the textbook Kelly is full Kelly, and it produces volatility. In practice, successful traders use half-Kelly (22.7%) or quarter-Kelly (11.35%) to reduce volatility while keeping most of the return advantage.

With quarter-Kelly on Setup A:

  • Risk per trade: $100,000 × 11.35% = $11,350
  • Position size: $11,350 / expected loss per trade = depends on your stop placement

This is still 11x larger than risking a fixed $1,000 per trade. But it's not dangerously leveraged, and it captures 80%+ of Kelly's return advantage with only 20-30% of Kelly's volatility.

For Setup B (weak edge):

f* = (0.55 × 1.111 - 0.45) / 1.111
f* = (0.611 - 0.45) / 1.111
f* = 0.161 / 1.111
f* = 0.145 = 14.5%

Quarter-Kelly: 3.6% of capital per trade. If your account is $100,000, risk $3,600 per trade.

Now you have a ratio: Setup A at 3.1x the size of Setup B (11.35% / 3.6%). This is how professionals size—not "one size fits all," but sizes proportional to edge.

Why Conservative Traders Under-Size: The Psychology​

Despite the mathematical advantage of sizing proportional to edge, most traders under-size their best ideas. Why?

Fear of ruin. "If Kelly says risk 45% per trade, I could be down 40% after a losing streak. What if I hit 10 losses in a row? I could blow up." This is a valid concern about full Kelly, but it's not a valid concern about quarter-Kelly or half-Kelly, which have drawdown profiles comparable to other risk-management strategies.

Recency bias. Your edge is historical (backtested on 500 trades). Your brain's recent experience might be different (you just had 3 losses). This makes the historical edge feel unreliable. So you size conservatively "until I'm sure." But "being sure" takes even more recent wins, and by then, you're past the setup that triggered the sizing decision.

Overconfidence paradox. Traders often exhibit two opposing biases:

  • They overestimate the accuracy of their short-term market predictions
  • They underestimate the reliability of their documented long-term edges

This combination leads to under-sizing edges (because they doubt long-term studies) while over-sizing low-conviction trades (because they're confident in current market direction).

Volatility fear. Quarter-Kelly sizing will produce larger individual trade wins and losses. If your standard trade is ±$1,000, a quarter-Kelly trade on a strong edge might be ±$5,000. The larger loss feels worse, even though the return-to-risk ratio is better. Humans don't optimize for return-to-risk; they optimize for loss aversion.

Portfolio theory confusion. Modern Portfolio Theory (MPT) often teaches "equal weighting across assets." This is reasonable for uncorrelated assets (stocks, bonds, commodities). But it's absurd for a single asset class (your trading setups) where you have measured edges. Within a single edge, variance and correlation are knowable. Use them.

Identifying Your True Edge (Before You Size It)​

Not all "edges" are real. Before you justify sizing up, you must verify:

  1. Statistical significance. Do you have enough samples? 500 trades minimum. If you have 50 trades at 68% win rate, that could easily be noise. With 500 trades at 68%, the edge is likely real. Use the binomial test: is the difference from 50% significant at p < 0.05? (Usually requires ~100 trades at 65% win rate.)

  2. Out-of-sample verification. Did you test on the same data you optimized on? (Over-fit.) Test on a separate period you didn't use for optimization. If the edge only works on your training set, don't size it.

  3. Macro-regime testing. Does the edge work across different environments? Test separately on:

    • Bull markets (2010-2019)
    • Bear markets (2008, 2022)
    • Range-bound markets (2015, 2016)
    • High-volatility periods (2020, 2023)
    • Low-volatility periods (2017)

    If the edge breaks down in any regime, you don't actually have an edge—you have a regime-specific setup. Size it accordingly (only during that regime).

  4. Cost adjustment. Your backtests assumed perfect fills and zero slippage. Real trading has slippage, commissions, and taxes. Is your edge large enough to survive these costs? If your edge is $386 per trade and your execution costs are $150, true edge is $236. Adjust position sizing accordingly.

  5. Correlation to portfolio. Do your high-edge trades correlate with your other holdings? If you're long equities and your best edge is also an equity long-only system, your portfolio is concentrated, not diversified. Consider diversifying before aggressively sizing the best edge.

Case Study: Under-Sizing in Practice​

A professional day trader backtested a high-frequency momentum strategy on ES (S&P 500 futures). Results:

  • 62% win rate
  • Avg win: $187 per contract
  • Avg loss: $128 per contract
  • 1.46:1 win-loss ratio
  • Out-of-sample verified across 2015-2022
  • Stable across bull, bear, and range-bound markets

Kelly calculation:

f* = (0.62 × 1.46 - 0.38) / 1.46
f* = (0.905 - 0.38) / 1.46
f* = 0.525 / 1.46
f* = 0.36 = 36%

Quarter-Kelly: 9% of capital. On a $200,000 account, risk $18,000 per trade. With ES at $50 per point, expected loss per trade is ~$1,000 (using a typical stop 20 points away). This justifies 18-contract positions.

Yet the trader's actual sizing:

  • Years 1-2: Started with 1 contract (risking $200/account)
  • Years 3-4: After consistent profits, increased to 2 contracts (risking $400/account)
  • Years 5-6: Slowly increased to 5 contracts (risking $1,000/account)

For the first four years, he was massively under-sized. His annualized return was 15%. If he'd sized at quarter-Kelly from year 1, expected annual return would have been 45-50%. The under-sizing cost him:

  • Years 1-2 lost profits: 4 × (50% - 15%) × $200,000 = $280,000
  • Years 3-4 lost profits: 4 × (50% - 20%) × $200,000 = $240,000
  • Total opportunity cost: $520,000

By the time he reached appropriate sizing (year 6), he'd already lost over half a million in compounding opportunity. Once his account had grown to $1M, quarter-Kelly sizing could finally be applied, but the early years of compounding had been squandered.

The Optimal Sizing Framework: Risk vs. Edge​

Here's a decision tree for position sizing:

If you have NO measured edge (or edge unverified):

  • Use fixed risk (e.g., 1% of account per trade)
  • All trades same size
  • Conservative is appropriate here; you're flying blind

If you have a WEAK edge (55-58% win rate, 1.0-1.2 risk-reward):

  • Use fixed risk (1% of account per trade)
  • OR use quarter-Kelly (3-5% of account per trade)
  • Optimize for Sharpe ratio, not absolute return

If you have a MODERATE edge (58-63% win rate, 1.2-1.6 risk-reward):

  • Use half-Kelly to quarter-Kelly (5-10% of account per trade)
  • Vary sizing within this range based on market regime
  • Optimize for Calmar ratio (return / max drawdown)

If you have a STRONG edge (>63% win rate, 1.6+ risk-reward):

  • Use quarter-Kelly to 1/8-Kelly (10-20% of account per trade)
  • Vary sizing based on market regime volatility
  • Optimize for compounding (return × years)

FAQ​

Isn't Kelly Criterion gambling advice? Shouldn't traders avoid it?​

No. Kelly was developed by a Bell Labs scientist and is used by institutional hedge funds, insurance companies, and professional traders worldwide. It's not gambling advice; it's optimization mathematics. The only "gambling" aspect is that it applies to any probabilistic decision with measurable edge—betting, trading, investing. The math is sound regardless of application.

What if my edge changes over time? Should I re-calculate Kelly?​

Yes. Quarterly or semi-annually, re-run your statistics on recent trades (last 100-200 trades). Recalculate win rate and win-loss ratio. If it's changed materially, update your Kelly sizing. If your edge has weakened from 65% to 58%, reduce sizing accordingly. This isn't market timing; it's honest assessment of current edge.

How do I know if my edge is really 68% or if I'm just seeing a lucky streak?​

Binomial confidence interval. With 500 trades at 68% win rate, the 95% confidence interval is roughly ±3%. So your true edge is likely 65-71%. With 50 trades at 68%, the confidence interval is ±13%—i.e., your true edge could be anywhere from 55% to 81%. You need sample size. Use the rule of thumb: 100 trades minimum at 65% win rate to claim an edge with reasonable confidence.

Does under-sizing ever make sense?​

Yes. If your edge is marginally significant (52% win rate, barely above 50% threshold), under-sizing is appropriate. The expected value is small, and larger sizes increase ruin risk. Also, if your edge only works in certain market regimes (FOMC days, earnings season), size it based on frequency of occurrence, not long-term average. If the setup only appears 5% of trading days, your "part of portfolio" sizing should be ~5%, not 20%.

How do I reconcile Kelly Criterion with risk management rules like "never risk more than 2% per trade"?​

Kelly is a position-sizing framework. "Never risk more than 2% per trade" is a risk-management cap. Use both: calculate optimal Kelly size, then cap it at 2% max account risk per trade. If Kelly says 5%, but your rule is max 2%, use 2%. The cap prevents ruin even if your edge assumptions are wrong.

If I size up my best ideas, won't I wipe out faster if they fail?​

Possibly, if your edge is misidentified. But if your edge is real and properly verified, sizing larger on best ideas actually reduces ruin risk because you're earning more per unit of volatility. You're increasing Sharpe ratio, which is the best insurance against ruin. The danger is sizing up on fake edges. So verify first (out-of-sample, across regimes), then size.

What's the relationship between Kelly sizing and drawdown?​

Quarter-Kelly produces maximum drawdowns roughly 1.5-2x the baseline fixed-risk drawdown. Full Kelly can produce drawdowns 3-4x larger. Half-Kelly is a reasonable compromise: 2x the drawdown for 3x the return. Plot your system's historical peak-to-trough drawdown under fixed sizing, then under Kelly, to understand the trade-off.

Summary​

Under-sizing your best ideas is a silent killer of long-term returns. A trader with a 68% win-rate edge sized identically to a 55% win-rate edge leaves 50%+ of potential returns on the table. Over a 10-year career, this costs hundreds of thousands of dollars.

The solution is Kelly Criterion sizing: a mathematical formula that determines optimal position size based on your win rate and win-loss ratio. Professional traders, hedge funds, and institutional investors use Kelly Criterion (or approximations of it) routinely. The textbook Kelly produces high volatility, but quarter-Kelly and half-Kelly provide 70-80% of Kelly's return advantage with only 25-50% of its drawdown.

The barrier to implementation isn't mathematical; it's psychological. Traders feel safer with equal sizing across all trades, even though the math proves this is suboptimal. They also fear the larger losses that come with proper sizing, even though proper sizing actually reduces risk-adjusted drawdowns.

If you've backtested a setup and verified the edge is real (out-of-sample, across multiple market regimes, with sufficient sample size), then sizing your position based solely on risk tolerance is leaving money on the table. Calculate your Kelly fraction. Multiply by 0.25 (for quarter-Kelly) or 0.5 (for half-Kelly). Use this as your position sizing target. Watch your compounding acceleration. The trade with the highest expected value should be your largest trade—not your smallest.

Next​

→ Abandoning a Sound System During Drawdown