Taleb: success "requires survival but not the other way around." Two of our futures strategies ended their backtests with solid profits. Both would have broken a prop firm's trailing drawdown on the way. Only the day-by-day equity curve showed it. I judge the path first.

Another red index day, another green short. Oct 22: Nasdaq-100 −2.90% close-to-close. My US100 CFD short: ~29,060 → ~28,890, about +170 points. MT5 chart unchanged. Still a single trade, nowhere near a track record. NDX source: chartexchange.com/symbol/ind…

The longer you hold SPY, the more lopsided its returns get. Skew of log returns since 1993: −0.25 by day, −0.74 by month, −0.92 by quarter. Day to day it is close to symmetric. Over quarters the bad side dominates. Descriptive, not a backtest.

The textbook RSI(2) setup adds a 200-day trend filter. On ten big US stocks since 2020 it earned less on 9 of 10. Meta 1.3x instead of 2.2x, Apple 1.25x instead of 1.6x. Only Tesla gained, 0.96x vs 0.94x. (That version also exits on the 5-day average.) Backtest.

I fixed the weights of my 20-rule portfolio on data to June 2023, then measured the three years after. 11.3% a year, Sharpe 1.73, worst drawdown 5.3%. SPY made 18.9% a year with an 18.8% drawdown. Backtest with costs. What it holds and the catch below.

If a backtest shows one cost level only, I read it as a pitch. We reran 49 rules from trading newsletters with our own cost model. 47 made money after costs. At triple costs, 43 still did. Triple costs is the first stress I run, before any Sharpe ratio. Backtests.

Volatility has memory. After SPY's calmest 20-day stretches, the next 20 days ran at 10% annualized volatility. After the wildest: 26%. Prices are hard to forecast. Risk much less so. Descriptive, SPY daily 1993 to 2026.

By late morning in New York on Thursday: +$4,247.12 on MNQ, across prop accounts that all copy one automated strategy. No clicks. The part screenshots skip: a copier copies the bad weeks too. In our simulation, 80 to 100% of copied accounts breached in the same week.

We split 20 of our rules into two halves of about three years. The 5 best of the first half fell from a Sharpe ratio of 1.45 to 1.07. The 5 worst rose from 0.39 to 0.77. Luck fades. Judge a rule on its full history. Backtests with costs.