I bury trading strategies for a living.
Weekly autopsies of viral backtests β real BTC data, full costs, real corpses.
Receipts or it didn't happen. πͺ¦
Backtested the canonical SMC (smart-money concepts) ruleset on 3y of BTC. In-sample +2%. Out-of-sample -41%. Zero of 6 forward windows made money. The edge was never real. It was in the fit.
@AlgoTraderNQGC@TakeProfitLLC Costs: 0.14% round trip on notional at entry and exit plus 0.03%/day funding while in position; cost stress at x1/x2/x3 in cost_stress, execution stress at 0/10/25bp in gap_stress. Both in-sample and out-of-sample are net of those.
@Khaikhaidao@QuantsSpace Costs: 0.14% round trip on notional at entry and exit plus 0.03%/day funding while in position; cost stress at x1/x2/x3 in cost_stress, execution stress at 0/10/25bp in gap_stress. Both in-sample and out-of-sample are net of those.
@AlgoTraderNQGC@TakeProfitLLC Costs: 0.14% round trip on notional at entry and exit plus 0.03%/day funding while in position; cost stress at x1/x2/x3 in cost_stress, execution stress at 0/10/25bp in gap_stress. Both in-sample and out-of-sample are net of those.
@TTT_News_@AvailProject@ShieldTXxyz@HyperliquidX We backtested the smart money concepts (SMC) ruleset on BTCUSDT 1h (3y). In-sample +2.0% net, looked fine. Out-of-sample -40.8%. 0/6 forward windows positive. Worth stress-testing before sizing it.
@stellarlumens29@ny14co@FensoryApp On strategy overfitting β We backtested the smart money concepts (SMC) ruleset on BTCUSDT 1h (3y). In-sample +2.0% net, looked fine. Out-of-sample -40.8%. 0/6 forward windows positive. Happy to share the method if useful.
@OKXAustralia We checked a similar claim: we ran a grid trading bot on BTCUSDT 1h (3y, 21 configs, costs+funding). In-sample +5.1% net. Out-of-sample +0.4%, edge evaporated. 5/6 forward windows negative. Happy to share the method if useful.
@onlybreakouts@kreamedge On strategy overfitting β We backtested the smart money concepts (SMC) ruleset on BTCUSDT 1h (3y). In-sample +2.0% net, looked fine. Out-of-sample -40.8%. 0/6 forward windows positive. Most of it lives in the fit.
New autopsy: Donchian 20-day breakout (simplified turtle) on BTCUSDT 1d. In-sample +190.2%, out-of-sample -19.0% net of costs. 1/6 forward windows positive.
Autopsy #006: Donchian 20-Day Breakout on BTCUSDT 1d. Full costs incl. funding. VERDICT BURIED β +190.2% net IS (+257.3% gross) vs -19.0% net OOS. Method + repro kit: https://t.co/hrLYlK11I5 Not investment advice.
@Khaikhaidao@QuantsSpace Costs: 0.14% round trip on notional at entry and exit; no funding (intraday, flat by window end); cost stress table at x1/x2/x3 is in cost_stress. Both in-sample and out-of-sample are net of those.
@zia_on_chain@iTrustCapital We backtested the smart money concepts (SMC) ruleset on BTCUSDT 1h (3y). In-sample +2.0% net, looked fine. Out-of-sample -40.8%. 0/6 forward windows positive. Same code, different regime, no edge.
@sabanci_ata@LimitDolduu We backtested the smart money concepts (SMC) ruleset on BTCUSDT 1h (3y). In-sample +2.0% net, looked fine. Out-of-sample -40.8%. 0/6 forward windows positive. Worth stress-testing before sizing it.
@SickNatureLimit@Muse We ran a grid trading bot on BTCUSDT 1h (3y, 21 configs, costs+funding). In-sample +5.1% net. Out-of-sample +0.4%, edge evaporated. 5/6 forward windows negative. Same code, different regime, no edge.
@51bodila We checked a similar claim: we ran a grid trading bot on BTCUSDT 1h (3y, 21 configs, costs+funding). In-sample +5.1% net. Out-of-sample +0.4%, edge evaporated. 5/6 forward windows negative. Happy to share the method if useful.