How Much Should You Bet? A Backtest of Position Sizing Methods

Position sizing is the decision that determines whether your trading strategy survives its first crisis. Get it wrong and edge doesn't matter — the wrong size turns a 0.8 Sharpe strategy into a blown account. The right size is the difference between a -33.7% drawdown and a -19.9% one, and it's the only variable in trading you fully control.

This article compares four position sizing approaches — buy-and-hold, a trend filter, volatility-targeting, and the combination of trend + vol-targeting — using 15 years of SPY data from our backtest engine. We also look at Kelly betting and half-Kelly as the theoretical benchmark for discrete trades.

What Is Position Sizing?

Position sizing is the decision of how much capital to allocate to each trade or strategy. It's distinct from asset selection or entry timing — it's the "how much" question that sits between your edge and your P&L.

Four common approaches span the spectrum from simple to systematic:

Fixed-Fractional Sizing

The simplest rule: risk a fixed percentage of your capital per trade. Ralph Vince's "Optimal f" method takes this to its data-driven extreme — it examines the historical sequence of trades and finds the fraction that would have maximized geometric growth. The limit: it's entirely backward-looking and can recommend aggressive sizing that fails the moment the future diverges from the past (Quantpedia). In practice, most traders use 0.5–2% per trade as a heuristic, which is conservative enough to survive bad streaks but leaves significant return on the table in strong markets.

The Kelly Criterion

The Kelly criterion comes from a Bell Labs engineer who realised the same math that described signal-noise problems also described betting — the fraction that maximises long-run geometric growth given a known edge: f* = p − (1−p)/b, where p is the win probability and b is the win/loss ratio (Wikipedia).

For a strategy with a 60% win rate at even odds, Kelly says bet 20% of capital. That's aggressive — and that's the problem. Full Kelly produces spectacular growth in theory and stomach-churning volatility in practice. The landmark Ziemba and Hausch (1986) simulation showed that over 700 bets with a 14% edge, full Kelly ended with a median wealth of $17,269 — but some paths hit $18, barely surviving. Half-Kelly had a much tighter range ($145–$111,770) and a higher median floor (Carta 2020).

In our own SMA 200 trend strategy on SPY (39 trades over 15 years), the Kelly fraction was 23.3%, but the small sample means this estimate has wide error bars — a reminder that Kelly requires reliable edge estimation, which is harder than it looks in live markets.

Volatility-Targeted Sizing

Instead of betting a fixed fraction, vol-targeting sizes positions inversely to recent volatility. The goal: keep portfolio risk constant regardless of market conditions. When volatility spikes, you shrink exposure. When it compresses, you ramp up (within limits).

The Concretum Group's comparison of vol-targeting frameworks found that volatility parity (equalizing each asset's vol contribution) produces smoother equity curves and lower drawdowns than fixed sizing, while adding pyramiding to vol-parity boosted returns at the cost of higher volatility. For our single-asset test, we targeted 15% annual volatility on SPY, scaling up to a maximum of 2x leverage when vol dropped below 7.5%.

The Backtest: Four Methods, One Asset, 15 Years

We ran all four methods on SPY from January 2010 through December 2025 using the research-service backtest engine, with a flat 1 bps cost assumption (SPY is cheap to trade). The trend filter uses a 200-day simple moving average — in when above, cash when below. Vol-targeting uses a 63-day trailing volatility window scaled to a 15% annual target.

Results at a Glance

What the Numbers Show

Buy & Hold wins on raw return — as it should during one of the strongest bull markets in history. The 13.9% CAGR is the ceiling. But you ride every drawdown, including the 2020 COVID crash and the 2022 bear market, for a maximum peak-to-trough loss of -33.7%.

The SMA 200 trend filter sacrifices 3.7 percentage points of CAGR but cuts volatility by a third and nearly halves max drawdown. The Sharpe ratio actually improves to 0.91 — the highest in the group. During sharp drawdowns, being in cash is a feature, not a bug. The cost: whipsaws in choppy markets and the emotional tax of getting back in after a false signal.

Vol-targeting on its own is a middle path — 12.9% CAGR (just 1% behind buy-and-hold) with max drawdown improved from -33.7% to -24.6%. The strategy captured most of the upside while trimming exposure during high-volatility periods like 2020 and 2022. The Sharpe ratio doesn't improve because scaling introduces its own timing risk, but the raw drawdown improvement is meaningful for anyone whose sleep quality correlates with portfolio volatility.

The combined SMA 200 + vol-targeted approach produces the gentlest equity curve of the four: -19.9% max drawdown and the second-lowest volatility at 13.9%. You give up more upside (11.5% CAGR vs 13.9%), but the ride is subjectively smooth — the kind of strategy that keeps you in the seat when instinct says sell everything.

How This Applies to Systematic Trading

At RiskHarvest, we spend most of our time on the right side of this table — the vol-targeted and filtered approaches. The reason is straightforward: our edge is in harvesting risk premia systematically, not in timing the next crash. Vol-targeting aligns position size with the one variable we can actually measure in real time (current market volatility), and trend filters handle the regime shifts that would otherwise blow through a static vol target.

The SMA 200 + vol-targeted combination is the closest proxy to what a systematic risk-premia strategy looks like in practice: you're in the market capturing the premium when conditions are normal, you scale down when vol spikes, and you step aside entirely when the trend breaks. That 11.5% CAGR with a -19.9% drawdown is not a target — it's a historical example of what that framework produced during a period that included two major drawdowns.

The lesson for anyone building or running a systematic strategy: ensure you design your position sizing mechanism not just your signal. The signal determines your edge; position sizing determines whether you survive long enough to collect it. Backtest them together, not separately — the interaction between entry logic and size logic is where most strategies either find their groove or fail.

Position sizing won't turn a losing strategy into a winning one. But it's the difference between a winning strategy that survives its first crisis and one that doesn't.

Educational content only. Historical backtest results do not guarantee future returns. All investments carry risk.