Overview & Experiments 55 Synthesis Roadmap Lookahead Audit

Vol-target sizing — invert position by predicted vol

Inconclusive

2026-05-20 vol-targetsizingrisk-managementbacktest

Hypothesis

Scaling each trade's size by clip(target_vol / trailing_RV_4h_ann, 0.5, 2.0) improves Sharpe by >= +0.2 with <= 15% loss of total return on the three currently-deployed strategies (BB_EXTREME_1, BB_EXTREME_2, DONCHIAN_50).

Verdict

**MARGINAL** — small Sharpe lift +0.080, DD reduction +0.1pp, but lift is under the +0.2 promote gate. Could re-test with LGBM predicted vol (Master-LGBM IC 0.81 > persistence 0.68) before deciding.

avg_pnl_ratio

+1.1186

avg_sharpe_lift

+0.0805

avg_dd_reduction

+0.0006

strategies_tested

['DONCHIAN_50', 'BB_EXTREME_1', 'BB_EXTREME_2']

target_vols_tested

[0.5, 0.6, 0.75]

best_target_vol_by_strategy

{'DONCHIAN_50': 0.5, 'BB_EXTREME_1': 0.413, 'BB_EXTREME_2': 0.404}

Vol-target sizing — invert position by predicted vol

2026-05-20 · status: inconclusive · 21.8s

Hypothesis: Scaling each trade's size by clip(target_vol / trailing_RV_4h_ann, 0.5, 2.0) improves Sharpe by >= +0.2 with <= 15% loss of total return on the three currently-deployed strategies (BB_EXTREME_1, BB_EXTREME_2, DONCHIAN_50).

Verdict: MARGINAL — small Sharpe lift +0.080, DD reduction +0.1pp, but lift is under the +0.2 promote gate. Could re-test with LGBM predicted vol (Master-LGBM IC 0.81 > persistence 0.68) before deciding.

Key metrics

metric	value
avg_sharpe_lift	`+0.0805`
avg_pnl_ratio	`+1.1186`
avg_dd_reduction	`+0.0006`
strategies_tested	`['DONCHIAN_50', 'BB_EXTREME_1', 'BB_EXTREME_2']`
target_vols_tested	`[0.5, 0.6, 0.75]`
best_target_vol_by_strategy	`{'BB_EXTREME_1': 0.413, 'BB_EXTREME_2': 0.404, 'DONCHIAN_50': 0.5}`

Approach

Backtest of currently-deployed strategies (DONCHIAN_50, BB_EXTREME_1, BB_EXTREME_2) over 2022-01-01 -> 2026-05-01 with run_backtest_live_sim. For each trade, trailing 4h realized vol (annualised) at entry_time is looked up; multiplier = clip(target_vol / predicted_vol, 0.5, 2.0). Each trade's pnl_usd, fees_usd, funding_usd scale linearly. Stops unchanged. Vol proxy = trailing 4h RV (causal, persistence-baseline; Master-LGBM would improve on this).

Per-strategy summary

strategy	target_vol	n_trades	baseline_pnl	baseline_sharpe	baseline_dd	scaled_pnl	scaled_sharpe	scaled_dd	sharpe_lift	pnl_ratio	dd_reduction	mean_mult
DONCHIAN_50	0.657	127	$7.57	0.476	0.4%	$7.43	0.478	0.3%	0.001	98%	+0.1pp	1.08x
DONCHIAN_50	0.5	127	$7.57	0.476	0.4%	$6.12	0.5	0.2%	0.024	81%	+0.2pp	0.85x
DONCHIAN_50	0.6	127	$7.57	0.476	0.4%	$6.93	0.484	0.3%	0.007	92%	+0.1pp	1.00x
DONCHIAN_50	0.75	127	$7.57	0.476	0.4%	$8.34	0.474	0.3%	-0.002	110%	+0.0pp	1.21x
BB_EXTREME_1	0.413	150	$20.45	0.845	0.2%	$25.19	0.942	0.2%	0.097	123%	-0.0pp	1.09x
BB_EXTREME_1	0.5	150	$20.45	0.845	0.2%	$30.30	0.941	0.3%	0.096	148%	-0.1pp	1.26x
BB_EXTREME_1	0.6	150	$20.45	0.845	0.2%	$33.97	0.935	0.3%	0.09	166%	-0.1pp	1.44x
BB_EXTREME_1	0.75	150	$20.45	0.845	0.2%	$37.74	0.917	0.4%	0.072	185%	-0.1pp	1.65x
BB_EXTREME_2	0.404	116	$27.86	1.129	0.2%	$36.66	1.25	0.2%	0.121	132%	+0.0pp	1.10x
BB_EXTREME_2	0.5	116	$27.86	1.129	0.2%	$44.51	1.247	0.2%	0.118	160%	+0.0pp	1.29x
BB_EXTREME_2	0.6	116	$27.86	1.129	0.2%	$48.45	1.238	0.2%	0.109	174%	-0.0pp	1.46x
BB_EXTREME_2	0.75	116	$27.86	1.129	0.2%	$52.57	1.216	0.3%	0.086	189%	-0.1pp	1.66x

Best target_vol per strategy

strategy	target_vol	sharpe_lift	pnl_ratio	dd_reduction
BB_EXTREME_1	0.413	0.097	123%	-0.0pp
BB_EXTREME_2	0.404	0.121	132%	+0.0pp
DONCHIAN_50	0.5	0.024	81%	+0.2pp

Avg across strategies (best-tv each): Sharpe lift +0.080, PnL ratio 112%, MaxDD reduction +0.1pp

comparison

Verdict

MARGINAL — small Sharpe lift +0.080, DD reduction +0.1pp, but lift is under the +0.2 promote gate. Could re-test with LGBM predicted vol (Master-LGBM IC 0.81 > persistence 0.68) before deciding.