The graveyard

Trading strategies that didn't work when I tested them

Every setup here is popular and easy to find on YouTube. Some I traded live on paper after they backtested well; others failed the backtest first. None of them made money. Each page has the exact rules I used, every trade, and why I think it failed.

Doesn't workTested live Fair Value Gap (FVG) -$5,240 273 live trades, 130 won Lost on 273 live SPX trades, and in backtests on SPY, QQQ and gold. Doesn't workTested live EMA crossover -$9,035 39 live trades, 27 won Won most of its 0DTE trades. The few losers went to zero. Doesn't workTested live ATR pullback -$7,580 55 live trades, 19 won Fading rallies made money. Buying dips lost three times as much. Doesn't workBacktest only ICT order blocks and the ICT model 52% of 520 order-block settings profitable: a coin flip Order blocks were a coin flip. The full sweep, shift and FVG model did no better than fake levels. Doesn't workBacktest only Opening range breakout (ORB) 11 of 12 15-minute ORB exit settings lost money on SPX 15, 30 and 60-minute ranges, close and retest entries. Trading the opposite side did as well. Doesn't workBacktest only Buying 0DTE straddles and strangles -$125 per trade, opening straddle, worse at every exit time At the open, at 09:45 and at noon. The market charged exactly for the movement it delivered. Doesn't workBacktest only Selling 0DTE iron condors 0 of 99 morning condor combinations profitable at real fills Fairly priced, so the bid/ask spread on four legs was the whole loss. Doesn't workBacktest only SPY scalping 0 of 70 scalping versions meaningfully profitable Momentum, VWAP, Bollinger, order book, round numbers, QQQ lead. The spread was bigger than every edge. Doesn't workBacktest + one live day Grid trading bot -$48,539 SPY and QQQ, 10-cent grid, 41 days Buy every dip, sell every rip, all day. Trend days wiped out weeks of small wins. Doesn't workBacktest only RSI(2) and calendar effects 76% vs 67% RSI(2) win rate vs random entries, same exit Plus turn of the month, FOMC days, options expiration week and IBS. None beat a random day. Doesn't workBacktest only SMT divergence and fair pricing -0.16R per SMT trade, vs -0.03R with no divergence Plus the 'open is fair price' playbook: 12 of 12 versions lost. Doesn't workBacktest only Fading back to the open 65% vs 62% return rate: the real open vs a fake one at 10:00 The open isn't a magnet. Five ways to fade back to it, none better than chance. Doesn't workBacktest only 0DTE credit spreads after a move +37% a trade in 45 days, then negative over 3 years With the move, against it, or far out of the money. The short sample flattered every version. Doesn't workBacktest only VIX regime switching 1.09 → 0.71 Sharpe ratio, as published vs fixed SPY, bonds and gold switched by VIX levels, with a machine-learning overlay. The edge was tomorrow's data.

Why publish the failures? Because they're the useful part. Anyone can show a strategy that worked on the months they picked. The setups that look convincing, get traded, and then lose are the ones that cost people real money, and the reasons they fail repeat: costs bigger than the edge, backtests fitted to their own data, and win rates that hide the size of the losses.

I may also have tested some of them wrong. Each page says how, and asks for your version.

"Tested live": paper trades on real market bid/ask quotes, one contract each, no commissions, updated 13 Oct 2026. "Backtest only": historical data with trading costs included.

This site records my own testing and trading. It isn't financial advice.

Disclosures · Privacy · [email protected]