I started options trading in 2020 during the pandemic: harvesting option premiums from option time decay. For 2 years I honed a system with specific entry, exit and adjustment criteria in a personal account. It performed really well, so well that I started a hedge fund and ran it from 2022 through 2024 on a similar strategy.
For the first year, performance was exceptional at 6-7% return per month, resulting in an over 100% (independently audited) annual return. We attracted about 50 LPs during that time.
It was even more remarkable because this first year covered the time period from Oct 2022 through Sept 2023 i.e. before the market started ripping up as it discovered the AI boom late 2023 (remember the S&P 500 was down ca. 20% in 2022 and most of the 2023 gains in the S&P were booked in the last 2 months of the year).
The strategy fit the market regime, the numbers confirmed it month after month, and I had little reason to question the design.
Then the regime shifted from late 2023. The same strategy, executed with the same discipline, stopped working.
That second year forced me back to the drawing board, and what I found there became the foundation of the OptionAgent AI Trading Co-Pilot, currently in pre-release at https://optionagent.ai. This is the case study, including the parts that were uncomfortable to live through.
What Went Right, and Why That Was the Problem
The original strategy was built for a specific market condition. When that condition held, returns were consistent. The fund looked healthy on every report.
Success like that hides a structural flaw. The strategy had one assumption baked into its core: that the high volatility regime it was designed for would persist. I had risk controls on individual positions, stop levels, sizing rules, all of it.
When volatility behavior, interest rate regime and market direction all changed in year two, every position suffered from the same cause. Position-level risk management was solving a problem one layer too shallow.
The Standard Fix, and Why I Rejected It
The common response to this kind of underperformance is to spread across more stocks. Add tickers, add sectors, reduce concentration. That is position diversification, and it is a reasonable tool for a different problem.
It did not fit mine. If every position runs the same strategy, and the strategy fails in the new regime, you have diversified the symptoms and kept the disease.
The question I ended up asking was a level deeper: what happens when the strategies hedge each other, instead of the positions?
Internal Hedging: Building Strategies That Offset Each Other
The system I rebuilt rests on a simple architectural principle. Each strategy in the book must behave differently across regimes, so that the condition that hurts one tends to support another.
In practice, that meant designing for three things:
Regime coverage. Every strategy gets mapped to the market conditions where it earns and where it bleeds. The portfolio needs strategies whose weak conditions do not overlap.
Offsetting exposures. Short-volatility income strategies get paired with structures that gain when volatility expands. Simultaneously run different strategies to benefit either from a range-bound market regime, and from a directional move. Monetize both at the index and opportunistically at the individual ticker level. Use different timeframes. Include a black swan tail hedge as insurance. The hedge lives inside the system, instead of being bolted on after a drawdown.
Risk defined at the architecture level. Sizing and limits apply to the interaction of strategies, in addition to individual trades.
This changed how I think about resilience. Protection stops being something you add when markets get scary. It becomes a property of the design itself, present before the shift arrives.
Why Automation Stopped Being Optional
Running one strategy by hand is manageable. Running several strategies that must stay balanced against each other, with risk limits enforced across all of them, exceeds what a person executes reliably day after day.
Three failure modes convinced me of this:
- Manual execution drifts. On stressful days, I deviated from my own rules.
- Rebalancing across strategies requires constant monitoring that humans sustain poorly.
- Risk controls only work when they trigger every single time, including the times you would rather they did not.
So we built the system into software. That became OptionAgent: a platform that runs multiple option strategies with internal hedging and enforces the risk framework automatically, with the same consistency on day 400 as on day 1.
The Lessons, Stated Plainly
If you trade a systematic approach, here is what my two years taught me:
Audit your assumptions, including in good years. A strategy that only works in one regime carries a hidden expiration date.
Diversify at the strategy level. More tickers under one strategy just concentrates regime risk while appearing to spread it. It's especially important in the current Goldilocks market that seems priced to perfection.
Design the hedge into the architecture. Hedges added during a drawdown arrive late and cost more.
Automate what must be consistent. Discipline encoded in software survives the days when your own discipline wavers.
The failure in year two cost me returns and some pride. It also produced a better system than the first year of success ever did. If your current approach depends on one market condition holding, now is the right time to examine it, before the regime does it for you.
Disclaimer: Past performance does not guarantee future results. This article is for informational purposes only and is not investment advice. Options trading involves substantial risk and is not suitable for all investors.