Energy Backtests Won on Win Rate and Still Lost to the S&P

Nothing in the news feed this weekend. No headlines, no earnings, no macro prints — which leaves the research slate to speak for itself, and this week it spoke mostly in caveats.

The Win-Rate Trap

Start with the two strategies that did the most trades. Buying COP at the close when it closed in the bottom 20% of its prior 20-session range returned +5.57% on $100,000 across 10 closed trades, with an 80% win rate. Over the same window, SPY buy-and-hold returned +76.34% — a trailing gap of 70.77 points. The best single trade made +3.55%; the worst lost -6.33%.

An 80% win rate that loses to a passive index isn't a paradox, it's arithmetic. The average win was smaller than the single worst loss, and ten trades is not many chances to recover from that. The XOM variant — buying after its daily news-sentiment score fell into the bottom quintile — ran 80 trades at a 51% win rate for +30.48%, still 45.86 points behind SPY, with a best trade of +7.11% and a worst of -3.80%. A coin flip with a better-shaped payoff, and still nowhere near the benchmark.

The Best of the Batch Was Second Place

The strongest result in the set came from the volatility filter: buying TTE at the close when Brent's 20-day realized volatility sat below the threshold, which returned +48.64% on $100,000 across 15 closed trades at a 67% win rate, best +8.61%, worst -4.56%. That is the one number in the batch that looks like a real edge in isolation. It also trailed SPY by 27.70 points.

Some of that gap is structural rather than damning — a rules-based approach that only holds when a condition triggers isn't the same exposure as buy-and-hold, and a benchmark printing +76.34% sets a brutal bar for anything that sits out part of the tape. But that's the honest frame: timing has to earn its cash drag, and on this evidence it hasn't.

Thin Samples, Confident Stories

Now the set-ups that sound smartest. Refined-margin momentum in MPC versus XLE: across 33 qualifying events, MPC beat XLE over the next 20 sessions in 54.5% of cases, mean excess return +0.49% (MPC +3.17% against XLE +2.68%). p=0.7030 — it doesn't clear the 5% bar, and against all eligible days the event edge was -1.22%, meaning the signal was worse than picking a day at random. Add overlapping forward windows and it's a non-finding wearing a finding's clothes.

The VLO-versus-XOM set-up after hot-CPI and weak-Brent anchors averaged 4.48 percentage points of outperformance over 20 sessions — across four occasions. Two of them carry it: +17.7% and +5.5% for VLO against XOM's +2.0% and -2.3%. The other two were a near-tie and a clear loss.

And the CVX free-cash-flow inflection in a falling Brent tape produced three qualifying quarters in roughly three years, in which CVX trailed XLE by an average of 1.09%, median -2.07%, beating the sector in just one — November 2024, +1.7%.

The pattern is consistent: the more narrative a set-up carries, the fewer observations it rests on. Thirty-three events at p=0.70, four CPI episodes, three quarters. That's a drawer of hints, not a library of edges. With no headlines to argue against it this weekend, the data leans one way — energy single-name timing has been a hard place to be clever, and the most confident-sounding set-ups deserve the most skepticism.