[Verification] Doubting My +66% Annual Returns — Bull-Market Tailwind, or Real Alpha?

The Urge to Earn More — and the Pause

The automated quant portfolio I’m running — a 50/50 mix of two systems trading different markets — backtests to +66% annual returns with a maximum drawdown (MDD) of −8.8% over five years. The drawdown is unusually shallow for that level of return.

Which naturally invites a thought. “If the drawdown is this small, isn’t there room to take on a little more risk — a bit more leverage — and pull the return up?” I wanted to find the sweet spot of the trade-off — giving up some MDD in exchange for CAGR.

But turning up leverage is, in the end, betting that this return holds going forward. Before scaling up, I needed to be suspicious of two things.

Doubt ① — Is this +66% just a “bull-market gift”?

2021–2025 was a sustained bull run in risk assets. If most of this return came from “the market just going up,” then leverage becomes poison the moment the market rolls over. Mistaking beta — riding a bull market — for alpha is one of the quant’s most common forms of self-deception.

How I tested it. I regressed the portfolio’s daily returns against a market benchmark and decomposed the return into two slices.

• Beta: the part earned by riding the market (and that gets cut down when the market falls)
• Alpha: the part earned regardless of what the market does (the actual skill)

Results.

Most tellingly, even during the actual 2022 bear market — the year risk assets got cut in half — this portfolio was up +22.7%. Across five years, not a single rolling 12-month window finished in the red.

Doubt ① resolved: this isn’t bull-market inflation — it’s market-neutral alpha.

Doubt ② — So the real danger is “alpha decay”

Alpha doesn’t die because the market falls. It dies because it wears down. When capital crowds into the same edge, the source of return gets compressed, and at some point it simply disappears. This is the quant’s real enemy.

I’d been assuming that my approach — continuously rotating what gets traded — offset this problem. But assuming isn’t verifying.

My portfolio is built out of about a dozen independent sleeves, each with a different character. I compared first-half vs. second-half risk-adjusted return (Sharpe) for each sleeve.

Initial result: Four sleeves — call them A, B, C, and D — looked noticeably weaker in the second half. “So some of them really are wearing down,” I thought.

Doubt ③ (self-check) — Real decay, or just “the recent market”?

Not stopping here is the whole point of this piece.

A simple first-half/second-half comparison has a trap. If an exceptionally strong early period gets bundled into “the first half,” even a perfectly healthy strategy will look like “the second half got worse.” You have to separate real decay from a recent regime that just happens to be hostile to that strategy.

So I broke the four sleeves out year by year.

A, B, and C were unambiguously regime effects. For all of them, the 2022 bear was the worst year, and they recovered afterward. The “looks weak” wasn’t decay — it was the market environment of that one year.

But Sleeve D — downward-trending and recently negative — kept my suspicion alive. This is exactly the kind of thing you can’t wave away.

The final test — skip the polluted metric; look at the signal’s source edge

Equity-curve-based Sharpe is a polluted metric — leverage, costs, and market regime are all baked in together. To see whether the alpha is alive or dead, you have to look at the strategy’s source-signal edge itself.

So I took the signal logic the bot actually uses and re-ran it as-is over the entire period. With leverage and costs stripped away, I broke the pure edge out by market regime — chop vs. trend.

(Side note: the original suspicion came from a calculation error in the tail of a time series that had been stitched together from multiple period-chunks — it had stamped a negative value where there shouldn’t have been one. When I reproduced it directly from the engine, the numbers came out clean. The lesson: don’t trust stitched-together summary data; reproduce from the source.)

Results.

• Steady year after year. 2025 and 2026 are just as healthy as the earlier years.
• Across all four cells of regime (chop/trend) × era (early/recent), every cell is positive. If anything, recent chop has been better than early chop.

Doubt ③ resolved: Sleeve D wasn’t decaying. The “weakness” I saw earlier was just an artifact of the stitched-together data.

Conclusion — Safe for now, and watched continuously

Of the four sleeves I scrutinized, zero were confirmed as structural alpha decay. All were either regime effects or stable. In other words, the portfolio that’s currently running is safe from alpha-decay risk — for now.

And this isn’t a check-once-and-move-on kind of thing. A decay monitor runs every day, automatically tracking how each sleeve’s return distribution shifts. If a distribution starts to drift statistically away from the backtest, an alarm fires — so I can catch alpha as it begins to wear down.

Three things I’m leaving with.

1. Doubt your returns — especially when they’re good. The moment you mistake beta for alpha, leverage becomes a weapon pointed at yourself.
2. “Looks weak” and “structurally dead” are not the same. If you can’t tell regime from decay, you’ll either throw away a perfectly healthy strategy or cling to a dead one.
3. Verify with the source signal, not summary metrics. A stitched-together curve will lie to you.

Only now do I have the right to come back to the original question — “How far do I turn up leverage to trade MDD for CAGR?” The return is real, and it’s still alive — I’ve confirmed that. The sweet-spot story is for the next post.

---

※ This post is meant to share the system’s reasoning process and verification methodology; specific signals, tickers, and position sizing are not disclosed. It is not investment advice.