The thing that finally made position sizing click for me was realizing I had been holding the wrong number constant. For years I sized in dollars. A fixed slug of capital per trade, same dollar amount whether the market was asleep or on fire. The dollars felt stable, so I assumed the risk was too. It was not. The risk was swinging around by a factor of three or four depending on the week, because a fixed dollar position in a quiet market and the same dollar position in a violent one are two completely different bets. I just could not see it because I was staring at the wrong variable.
Volatility targeting fixes that by holding risk constant instead of dollars. You pick a level of portfolio volatility you actually want to live with, you measure how volatile the market has recently been, and you size your gross exposure so those two things match. When the market gets choppier, your realized vol reading goes up, so you scale exposure down. When it calms, you scale back up. The dollar size floats. The risk stays put. That is the whole idea, and the reason I like it is that it requires you to forecast absolutely nothing. You are not predicting the next move. You are just reacting to how big the recent moves have been.
Why constant risk beats constant size
Volatility is one of the few things in markets that actually persists. Returns are close to a coin flip day to day, but volatility clusters. Calm follows calm, and once things get violent they tend to stay violent for a while before settling. That is not a theory I am selling you, it is one of the most robust empirical features of financial time series, and it holds especially well in crypto where regimes are loud and obvious.
That persistence is the entire edge here. If today was a wild day, tomorrow is more likely than usual to be wild too. So when your realized vol reading spikes, cutting exposure is not a guess about direction. It is a bet that the choppiness will linger, and that bet pays off often enough to matter. You end up mechanically pulling risk off the table as a storm is forming, before the worst of it lands, without ever having called the top.
The payoff shows up in the shape of your equity curve. A constant-dollar book takes its biggest hits precisely when vol is highest, because that is when a fixed position does the most damage. A vol-targeted book has already shrunk that position by the time the big down days arrive. Historically that tends to trim the fat left tail, smooth the ride, and improve risk-adjusted return, which is really just a fancy way of saying you get punched in the face less often and less hard.
A monthly rule set for a crypto book
Here is a version simple enough to run by hand once a month. I am using a crypto book as the example because that is where the vol swings are dramatic enough to make the whole thing worth the trouble.
- Pick a target volatility. Decide what annualized volatility you want the whole book to run at. Something in the neighborhood of 15 to 25 percent annualized is a reasonable, livable range for most people. This is your one real preference input, so pick a number you would be comfortable seeing on a rough month.
- Measure realized volatility. Take daily returns over a lookback window, compute their standard deviation, and annualize it by multiplying by the square root of the number of trading periods in a year. For crypto, which trades every day, that scaling factor is the square root of 365. This gives you the market's current realized vol.
- Compute the scaling factor. Divide your target vol by the realized vol you just measured. If you want 20 percent and the market is currently running at 40 percent, your factor is 0.5, so you run at half exposure. If realized vol drops to 20 percent, the factor is 1.0 and you run fully invested. When the market is unusually quiet, the factor can exceed 1.0, which implies leverage.
- Apply a leverage cap. Do not let that factor run wild. Cap it at something you can stomach, maybe 1.5 or 2.0 for a crypto book, so a stretch of eerie calm does not talk you into ten times leverage right before things break.
- Rebalance to target. Multiply your base position sizes by the capped factor and adjust holdings to match. Then wait until next month.
That is the loop. Measure, scale, cap, rebalance, wait. Nothing in there asks you where the market is going.
The lookback is the whole decision
Almost every meaningful choice in this system collapses into one question. How long a lookback do you use to measure realized vol.
A short window, say 20 trading days, reacts fast. It catches a vol spike quickly and de-risks you early, which is exactly what you want when a regime is genuinely turning. The cost is that it is twitchy. A single ugly day can jerk your exposure around and hand you a stack of trades that turn out to be noise. A long window, say 60 or 90 days, is calmer and cheaper to trade, but it is slow on the uptake, so it can leave you carrying too much risk well into a selloff before it finally notices.
I lean toward something in the middle, roughly a 30 to 60 day window, and if I am being honest a lot of people get a better result by blending a short and a long estimate rather than agonizing over the perfect single number. The blend reacts to real spikes without flinching at every twitch. Do not overfit this. The difference between a 40-day and a 45-day lookback is almost certainly noise, and if your backtest says otherwise, your backtest is lying to you.
Rebalance triggers and the failure modes
Rebalancing on a fixed monthly calendar is fine, and its blessing is that it forces discipline. But a pure calendar can leave you badly mis-sized when vol explodes mid-month. The fix I like is a no-trade band around the calendar. Rebalance on your scheduled date, and also rebalance off-schedule if the required scaling factor has drifted more than, say, 20 percent from where you last set it. That keeps your turnover low in normal times and lets you react fast when a regime actually breaks, without turning you into a day trader.
A few failure modes are worth naming, because I have walked into most of them.
- Trend and vol arriving together. Sometimes a market rips higher and gets more volatile at the same time. Vol targeting will trim your exposure into that strength, and you will feel like the system is fighting you. It is doing its job. It is buying you a smoother ride, and smoother rides cost some upside.
- Vol is not risk. A slow, quiet grind lower reads as low volatility right up until it does not. Standard deviation does not know the difference between calm-up and calm-down. Volatility targeting manages the size of your swings, not the direction, and it is not a substitute for a stop or a thesis.
- Chasing the calm. The single most dangerous moment is when realized vol drops to a multi-month low and your scaling factor is screaming at you to lever up. That is exactly when a leverage cap earns its keep. Honor the cap.
- Turnover bleed. Too short a lookback plus no rebalance band equals constant fiddling and fees that quietly eat the benefit. If you are trading every few days, your window is too twitchy.
None of this needs a model that predicts anything, which is the part I keep coming back to. You measure how rough the recent water has been, you carry less sail when it is rough and more when it is calm, and you cap how much sail you will ever fly. Run it for a few months, watch your worst days get less ugly than the market's worst days, and the appeal stops being theoretical.