Fixed fractional position sizing (e.g., "risk 1% per trade") is not risk management—it is risk theater that gives system

Fixed fractional position sizing is not risk management. It is a ritual that lets traders sleep at night while their capital bleeds out in exactly the scenario they swore they were protected against. I say this as someone who built and ran systematic strategies that survived multiple regime shifts.

Not because I was smarter than the math. Because I learned the hard way that the math, as commonly applied, is wrong in the moments it matters most. The standard playbook is seductive.

Risk 1% of equity per trade. Calculate position size from stop distance. Apply consistently.

The logic feels airtight: no single trade can kill you, drawdowns stay manageable, and over a long enough sample, edge compounds. This is what Van Tharp taught. It is what most prop trading education still teaches.

And in a stationary return environment with independent outcomes, it works exactly as advertised. We do not trade in stationary return environments with independent outcomes. The core failure of fixed fractional sizing is an assumption so deeply embedded that most practitioners never examine it.

The model treats each trade as a coin flip drawn from a stable distribution. Trade 47 knows nothing about trade 46. Your 1% risk on Friday is structurally identical to your 1% risk on Monday.

This is the independence assumption, and it is the load bearing wall of the entire framework. Pull that wall out and the building collapses. Here is what actually happens during regime shifts.

Volatility clusters. That is not an opinion. It is a well documented statistical property formalized by Mandelbrot in the 1960s and later captured in GARCH models.

When volatility clusters, losses cluster. Your system generates a string of losers not because the edge disappeared permanently but because the distribution of returns shifted underneath you while your position sizing model kept behaving as if nothing changed. Each individual trade risked "only" 1%.

But five consecutive losses in a compressed time window, each sized at 1% of a declining balance, compound into a drawdown whose probability your independent trade model dramatically underestimates. Let me put numbers to this. Under true independence, five consecutive losses at 1% risk each produce roughly a 4.9% drawdown.

Painful but survivable. Now introduce serial correlation of 0.3 across trade outcomes, which is conservative for trend following systems in a regime break. The probability of that five loss streak roughly doubles.

But the real danger is not the streak itself. It is that the streak tends to occur precisely when realized volatility is exploding, which means your stop distances are widening, which means your "1% risk" is actually being calculated against a noisier price path, which means your fills are worse, which means your true risk per trade is 1.4% or 1.7% while you still believe it is 1%. Stack that over five or eight trades and you are staring at a drawdown that was supposed to happen once in decades but just happened in nine trading days.

Ralph Vince's optimal f framework tried to address the sizing question more rigorously by maximizing terminal wealth. But it did not solve this problem either. Optimal f assumes a known and stable return distribution.

It will oversize you into the teeth of a regime change because it is optimizing against a historical sample that no longer describes the present. Kelly criterion suffers from the same structural weakness. Both frameworks answer the question "given this distribution, what is the right bet size?" Neither asks "is this still the right distribution?" The only honest position sizing model I have found worth running in production does not anchor to per trade risk at all.

It anchors to portfolio level drawdown velocity. The distinction matters enormously. Per trade risk is a bottom up calculation that ignores context.

Drawdown velocity is a top down measurement that captures regime information in real time. If your portfolio is declining at a rate that exceeds the expected drawdown path given your strategy's historical profile, that is information. Not noise.

Information. And the correct response is to reduce exposure as a function of how fast you are losing, not how much you are risking on the next individual trade. This means position sizing becomes adaptive.

Not in a discretionary, gut feel sense. In a systematic, measurable sense. You define drawdown velocity thresholds.

You map them to exposure multipliers. You test the entire framework across regime transitions, not just across the full backtest sample where everything averages out and looks comfortable. I built Vhalanx Core's infrastructure around this principle because I watched too many well designed strategies die from position sizing models that performed beautifully in backtests and catastrophically in the six days that actually determined whether the fund survived.

The strategy was fine. The edge was real. The sizing framework was a fairy tale that assumed tomorrow's losses would be politely uncorrelated with today's.

Fixed fractional sizing is not wrong in calm markets. It is dangerously incomplete in the markets that actually threaten your capital. And if your risk framework only works when you do not need it, you do not have a risk framework.

You have a comfort mechanism. I am curious how other systematic practitioners handle this. If you run live strategies through regime shifts, what does your sizing model actually anchor to when the distribution under you starts to move?