Sizing a Strategy With No Live Track Record

Every strategy is new once, and at that moment its only evidence is a backtest — a simulated history that, however carefully built, has never had to survive contact with real capital, real execution, and the version of the future that hasn't happened yet. Deciding how much money to give a strategy in this position is one of the genuinely hard judgment calls in running a fund, because there's a real tension in either direction: allocate too little and a genuinely good strategy never gets the capital it deserves quickly enough to matter; allocate too much and a strategy whose backtest was flattered by overfitting or good luck can do real, avoidable damage before anyone notices something is wrong.

The starting point is recognizing that a backtested Sharpe ratio systematically overstates what a strategy will do live, for reasons that have nothing to do with dishonesty — a certain amount of overfitting creeps in from testing many variations of an idea before settling on the one that looked best, and backtests are rarely charged the full realistic cost of trading at scale. A useful discipline is deliberately discounting the backtested Sharpe ratio by some margin before using it to size anything — treating, say, a backtested Sharpe of 1.5 as if it were closer to 1.0 for sizing purposes — precisely because a strategy sized as if the backtest were exactly right leaves no margin for the gap that almost always shows up between simulated and live performance.

A second input is simply how much the fund can afford to lose on this specific bet without it mattering to the business. A new, unproven strategy should never be sized so large that a plausible bad outcome — a losing streak well within what the backtest's own historical volatility would suggest is normal — would force the firm to shut it down for business reasons rather than investment reasons. Starting small enough that even an unlucky first few months is a tolerable, expected cost of doing this kind of research, rather than an existential threat, is what buys the strategy enough runway to actually prove itself over a meaningful live sample.

For example, a new strategy backtests at an annualized Sharpe ratio of 1.8 with a modest 8% annualized volatility. Rather than allocating capital as though a Sharpe of 1.8 were guaranteed to continue, a fund might size the initial live allocation using a discounted Sharpe assumption of around 1.0-1.2, and separately cap the dollar allocation so that a full year of returns at the strategy's own historical volatility, running unluckily, would still be a loss the fund can absorb without disrupting anything else it does.

What this means in practice

There's no formula that removes the judgment call entirely, but the discipline of discounting the backtested Sharpe ratio before sizing, and separately checking that a plausible bad outcome at the proposed size is tolerable, keeps a fund from accidentally treating a strategy's best-case backtested history as its expected live performance. The initial allocation is meant to be a starting point on a ramp, not a final answer — it should be revisited on a schedule as real live data accumulates, rather than left at its launch size indefinitely.