Volatility targeting: less drawdown or just less exposure?

The moving-average crossover study left EMA 9/21 on 4h candles with a median max drawdown of 75.16% over six coins. The textbook remedy is volatility targeting: instead of putting the same share of the account into every trade, size each trade so that an ordinary day's move costs the account the same share — a large position while the coin is calm, a small one while it is wild. On Janorium it is a sizing mode, VOL_TARGET.

Any rule that holds less draws down less, so the question worth asking is whether volatility sizing does better than simply holding less. This study takes the rule apart — the trend filter and the sizing one at a time — then sets volatility sizing against a fixed size with the same volatility, year by year and on ten coins it was not picked on.

Data and method

• Rule: long while EMA 9 is above EMA 21 on 4h candles, flat at the next open after it falls below — strategy #78 from the crossover study. No stop.
• Market and data: Binance spot, long only, BTC, ETH, BNB, XRP, ADA and DOGE, 2020-01-01 → 2026-10-01, 10,000 USDT per run. Fee 0.1% and slippage 0.05% on each side, fills at the next bar's open.
• Fixed size: 95% of equity per trade, as in the crossover study.
• Volatility target: sizing.mode: VOL_TARGET with targetVolPct: 2, atrRef: ATR(14) on daily candles and maxNotionalPct: 95. The position, in coins, is equity × 2% ÷ the daily ATR, so one average daily range moves the account by about 2%; spot has no leverage, so it is capped at 95% of equity. On BTC the position averaged about 56% of equity over 195 trades, between 21% and 95%; on DOGE about 31% (computed from the runs' trade lists).
• Trend filter: enter only while the 4h close is above the 50-day SMA on daily candles. The exit stays the EMA cross.
• Variants: the four combinations of sizing and filter — strategies #157 (the #78 rule unchanged), #158, #159 and #93, which makes both changes — plus, for each coin, a fixed size chosen so that its annualized volatility matches #93's on that coin (strategies #162–#172).
• How much was tried: the filter and the 2% target come from an earlier search of about 90 variants on these six coins — trend filters, whipsaw filters, stops, sizing rules and equity guards. This study added 42 configurations of the same rule (the sizes and targets below, plus one market filter), run as 444 dry runs and 50 stored runs. Ten more coins — LTC, LINK, TRX, BCH, ETC, XLM, DOT, UNI, AAVE and THETA — were kept aside and used only for the check at the end.
• Fresh runs: every stored run here was made on 2026-10-30. #93's runs reproduce the ones stored for it on 2026-10-21 to the last digit — on BTC +536.54% with a 24.82% max drawdown, then and now — and the fixed-95% runs repeat #78's from 2026-10-20 (BTC +351.63%). Nothing moved.

What each metric means: backtest metrics explained.

Smaller drawdowns, smaller returns

Volatility sizing on its own lowered the max drawdown on all six coins and raised the Sharpe ratio on five of them — but the return was higher on only two, ETH and XRP. The trend filter on its own kept the full size, lowered the drawdown on 5 of 6 and left it above 65% on XRP, ADA and DOGE. With both changes, #93 had the shallowest median drawdown, 25.96%, and a smaller drawdown than the fixed 95% on every coin; its return was higher on BTC and ETH only.

Each return links to its run. Volatility sizing barely changes when the rule trades — 318 trades on BTC either way — only how much each trade holds.

Backtest

EMA 9/21 crossover, 4h, 95% of equity (vol-targeting study: baseline)

BTCUSDT · 4h · Spot · 2020-01-01 → 2026-10-01

Return

+351.63%

Max drawdown

64.42%

Sharpe

0.76

Win rate

30.2%

Trades

318

Open the report →

Backtest

EMA 9/21 crossover, 4h, volatility-targeted size (vol-targeting study)

BTCUSDT · 4h · Spot · 2020-01-01 → 2026-10-01

Return

+326.29%

Max drawdown

36.26%

Sharpe

1.10

Win rate

30.2%

Trades

318

Open the report →

Is it only a smaller position?

On BTC the volatility target held about 56% of the account on average instead of 95%, and a smaller position alone lowers the drawdown. So the fair comparison is with a fixed size that carries the same risk. Two versions of it:

The same volatility, coin by coin. For each coin a fixed size was chosen so that the annualized volatility of the run's equity curve matched #93's on that coin — 54% of equity on BTC (volatility 19.94% against 19.93%), 27% on DOGE. This favours the fixed size: it is picked with hindsight from the whole period, while the volatility target only ever knows the last 14 days.

At the same volatility, the volatility-sized rule earned more and fell less on all six coins, and its Sharpe ratio was higher on all six too: median +337.20% with a 25.96% drawdown, against +264.11% with 27.93%. On BTC the fixed 54% is almost #93's own average position (about 56%) — the same average bet, placed differently: more in calm stretches, less in wild ones.

Backtest

EMA 9/21 crossover, 4h + daily SMA50 filter + volatility-targeted size (low-drawdown #78)

BTCUSDT · 4h · Spot · 2020-01-01 → 2026-10-01

Return

+536.54%

Max drawdown

24.82%

Sharpe

1.47

Win rate

36.4%

Trades

195

Open the report →

Backtest

EMA 9/21 crossover, 4h + daily SMA50 filter, fixed 54% of equity (vol-targeting study)

BTCUSDT · 4h · Spot · 2020-01-01 → 2026-10-01

Return

+409.47%

Max drawdown

28.04%

Sharpe

1.31

Win rate

36.4%

Trades

195

Open the report →

One size for all six coins. The target is a dial, and so is the fixed size. Turned through a range on the filtered rule, they trace out two lines (dry runs; the 2% and 95% rows equal the stored runs above):

A fixed 46% matched #93's median return (+331.02% against +337.20%) with a median drawdown of 35.63% instead of 25.96%; a fixed 28% matched its drawdown (26.29%) and made about half the return, +162.78%. Higher up the dial it is the same: a 4% target and a fixed 80% ended level on return (+792.80% and +801.64%), 44.06% against 50.83% on drawdown. What the target could not do on spot is reach the fixed 95%'s +1,072.81%: capped at 95% of equity, it can only shrink a position, never grow it past that.

Without the trend filter the picture holds, a little weaker (dry runs): a fixed 43% matched the volatility target's median return (+346.78% against +346.97%) with a drawdown of 43.71% against 35.30%, and against a fixed size of the same volatility per coin the target earned more on 6 of 6 but fell less on 4 of 6.

Year by year

The same variants were re-run on four windows, each starting with 10,000 USDT, the last ending on 2026-10-01 (dry runs). The vol-matched fixed sizes are the ones from the whole period.

Volatility sizing gave up most of the 2020–2021 boom — a median +144.89% for #93 against +847.10% at a fixed 95% — and lost a third as much in 2022. In 2025–2026 #93 made money on all six coins while the fixed 95% lost on all six.

Against the vol-matched fixed size the windows show how the target works. In 2020–2021 the fixed size earned more on 5 of 6 coins and #93 fell less on all 6; in 2022, 2023–2024 and 2025–2026 #93 ended ahead on all 6. A fixed size is the same in a boom and in a lull; the target shrinks the bet when a coin turns wild and grows it when the coin calms down, so it moves risk out of the wild years into the quiet ones. Its Sharpe ratio was the higher one on 5, 4, 6 and 6 of 6 coins in the four windows.

A market-wide filter on top — trade only while BTC closes above its 200-day SMA, strategy #94 — kept the rule out of 2022 entirely (no trades) and ended the whole period level with #93 (median +333.69% with a 25.58% drawdown, dry runs): what it saved in 2022 it gave back in 2020–2021, where it made a median +101.07%.

Ten coins it was not picked on

On the ten coins left aside, #93 made a median +97.89% with a 35.94% drawdown and made money on 9 of 10 — the same as when it was first checked on 2026-10-21. The unchanged #78 rule at a fixed 95% made a median +217.58% with a 88.77% drawdown (dry runs): #93 fell less on all 10 coins and earned more on 3. The volatility target without the filter made money on all 10 (median +87.31%, drawdown 38.87%, dry runs).

The fair test came out weaker here. Against a fixed size with the same volatility, #93 earned more on 8 of 10 coins and had the higher Sharpe ratio on 8, but the smaller drawdown on only 4 — LINK, XLM, AAVE and THETA. Medians: +97.89% with 35.94% against +82.70% with 37.08%. The return per unit of risk carried over to new coins; the drawdown advantage did not, not reliably.

Conclusion

• Volatility sizing cut the drawdown of the 4h EMA 9/21 crossover on every coin tested — 6 of 6, and 10 of 10 on the holdout coins, against a fixed 95% — mostly by holding less, and the return fell with it.
• At equal volatility it still earned more on 14 of 16 coins. It fell less on all 6 coins it was tuned on, but on only 4 of the 10 it was not.
• The target is a risk dial, not free return: 1% gave a median drawdown of 15.51%, 4% gave 44.06%, and on spot it cannot reach the raw return of a 95% position.

The rule itself and its fast and slow cousins: which moving-average crossover works best. How to read the cards: backtest metrics explained.