I spent the last week rebuilding a strategy I stopped trading in 2025.
The goal was not to rescue it.
The goal was to find out whether the original idea contained anything worth keeping, and whether I could trust the evidence enough to make that decision.
The first redesign looked dramatically better.
It was mostly taking more leveraged-equity risk.
One signal. Two different questions.
Corrfilter was a risk-control rule for a portfolio of leveraged equity funds, long-term Treasuries, and short-term Treasuries.
Its correlation-pressure statistic was supposed to decide when the portfolio should reduce risk.
The problem was that one number was answering two questions:
Are the leveraged equity holdings becoming one concentrated trade?
Is the Treasury position behaving like a useful hedge?
During early 2020, equity concentration was high while the Treasury hedge was working. The old statistic averaged those facts together. Strong hedge behavior made elevated equity risk look safer.
The portfolio followed the signal correctly.
The signal cut risk late, restored it while equity stress remained high, then cut again near the bottom.
The code did what it was told.
But a working Treasury hedge should not erase an equity-risk warning.
The redesign looked great
I separated equity stress from Treasury hedge quality and finalized six rules for translating equity stress into portfolio exposure before looking at their 2022 results.
One redesign returned 440.90% in the 2018 through 2021 development period, compared with 216.03% under the matching legacy architecture.
Risk-adjusted performance improved too. That made the redesign especially compelling.
It also increased average leveraged-equity weight from 42% to 71% and nearly doubled maximum drawdown.
Then I locked the rules and opened 2022.
All six redesigned versions lost more than their matching legacy versions.
All six rules got rekt.
The failure was not mysterious after I inspected the allocations. Approximately 22 percentage points moved from SHY, the short-term Treasury position, into UPRO and TQQQ.
The formula preserved the 56.4% figure.
It changed that figure from a combined equity-and-Treasury allocation into a leveraged-equity allocation.
The repair helped. The strategy still failed.
The corrected rule kept the original equity proposal unchanged. Hedge quality could reduce only the proposed TMF position, and every rejected dollar moved to SHY.
Every corrected version improved.
The best corrected rule still trailed SPY by approximately 21 percentage points in the 2022 out-of-sample test.
No candidate earned access to the unopened 2023 through mid-2025 true holdout.
That was the actual success:
Audit the evidence. Isolate the mechanism. Gate the final data.
The pipeline advanced.
Corrfilter did not.
If you have ever had a backtest improve after a refactor and could not explain exactly why, reply and tell me what changed. Signal timing? Portfolio construction? Execution? Accounting?
Those are usually more interesting questions than the new Sharpe ratio.
Disclosure: This is educational research-process review only. It is not personalized investment, trading, legal, tax, or financial advice.
Best,
Brian Christopher, CFA
BlackArbs LLC