One idea at a time: the test plan, the costs, the result and the verdict. Most of these failed. That's the point.
Autopsy 01Indices 10 years
360 breakout setups. 0 passed.
VerdictNot supported
I tested 360 versions of the opening-range breakout on the DAX, Euro Stoxx 50, FTSE 100, Nikkei and Russell 2000. That's 10 years of 1-minute data, with the spread paid on every trade.
I wrote down the pass mark before running anything. Not one variant reached it.
The part worth remembering: the best variant in hindsight would have made money. Picking the best one each year using only past data lost money on all five markets. Hindsight is the most profitable strategy there is, and nobody can trade it.
Next time someone shows you a backtest, ask how many versions they tried.
Variants tested
360
Variants with t ≥ 2
0
Best t (DAX 40)
1.66
Best t expected from luck, 72 tries
2.92
Walk-forward result per trade, risking $100
DAX 40−$0.51
Euro Stoxx 50−$12.12
FTSE 100−$6.15
Nikkei 225−$4.37
Russell 2000−$4.44
On a $10,000 account risking 1% a trade, the DAX's best variant picked in hindsight made +$5,167. Picked each year from past data only, it made −$990. Backtest, spread included. See all 360 variants plotted
Autopsy 02Gold, Nasdaq, S&P 500
+38% in one year. −19% on the eight before.
VerdictNot supported
My own trading bot made +38% in a one-year backtest on gold, Nasdaq and the S&P 500.
Then I ran the exact same rules on October 2016 to August 2024: −19.2% over 971 trades, with a 26% drawdown.
The catch: I picked its settings using 2024–26 data. A backtest on the data you tuned it on will nearly always look good. The only fair test is on data it has never seen.
That's what "out of sample" means, and why I named this series after it.
Oct 2025 to Sep 2026
+38.2%
Sep 2024 to Sep 2025
+41.0%
Oct 2016 to Aug 2024
−19.2%
Trades, 2016 to 2024
971
Oct 2016 to Aug 2024 (out of sample), net result by market from $100,000
One market made money on the unseen years and two lost; gold lost most. Three separate backtests of Aurora v12, each from $100,000, commission and overnight financing included. Maximum drawdown 2016 to 2024: −26%. Hover a bar for trades and win rate. Watch the chart zoom out
Autopsy 03S&P 500, gold 10 years
The edge was smaller than the spread.
Lesson
Intraday momentum is a published effect (Baltussen et al., 2021): if the S&P is above yesterday's close at 3:30 pm New York time, the last half hour tends to keep going.
I tested it on 10 years of S&P 500 and gold prices. Before costs, the S&P version made +0.77 basis points per trade. One spread costs 1.06. After real bid/ask costs: −0.55 bp per trade over 2,530 trades.
Costs aren't a detail you add at the end. They decide whether a strategy exists.
S&P 500, basis points per trade
Before costs+0.77
One spread1.06
After costs−0.55
Gold version: +0.13 bp before costs, a 1.86 bp spread, −2.12 bp net over 2,532 trades. Two other published effects, the overnight drift and the turn of the month, were not supported in my tests either. Backtests.
Autopsy 04S&P 500 forward test
Win rate is not the edge.
Forward test running
Random entries won 39% of the time. My breakout rule won 41%. That gap is noise.
What differed was how far winners ran: the rule hit its 3R target 13.7% of the time, random entries 8.7%.
Win rate alone tells you nothing. Your breakeven win rate is 1 ÷ (1 + reward/risk). With a 3:1 target that's 25%, before costs.
Honest footnote: this rule was picked from 72 variants. It's now in a forward test with no changes, and there's no verdict yet. I'll post the result either way.
Win rate, rule vs random
41.4% / 39.1%
Difference, p-value
0.11
Hit 3R, rule vs random
13.7% / 8.7%
Difference, p-value
< 0.001
Breakout ruleRandom entries, same stops and targets
Win rate, rule41.4%
Win rate, random39.1%
Hit 3R, rule13.7%
Hit 3R, random8.7%
S&P 500 opening-range breakout against random entries with the same stops and targets. Backtest. Forward test running, no verdict until it has enough trades.
GlossaryPlain words
The words these autopsies use
R
The amount you risk on one trade. A trade that makes twice what it risked is +2R.
t-statistic
How far an average result sits from zero, measured in units of its own noise. About 2 is the usual bar for one idea; trying many ideas raises the bar.
Basis point (bp)
One hundredth of one per cent: 0.01%.
Spread
The gap between the price you can buy at and the price you can sell at. You pay it on every trade.
Drawdown
The fall from the account's highest point to its lowest point before it recovers.
Out of sample
Data that played no part in choosing a strategy's rules or settings. Results on it are the fair test.
Walk-forward
Choose settings using only past data, trade the next period, move forward and repeat. It copies what you could really have done.
Breakeven win rate
The win rate at which a strategy makes nothing before costs: 1 ÷ (1 + reward/risk). With a 3:1 target, 25%.