March 2020: What the System Saw

The View from the Peak

On February 19, 2020, the S&P 500 closed at an all-time high. The VIX, Wall Street's measure of expected volatility, sat at 14 -- a number so low it suggested the market had collectively decided that nothing surprising would ever happen again. Unemployment was at historic lows. Earnings were beating expectations. Cable news discussed whether the bull run could last another decade.

The system was fully allocated. Its regime detector read "bull" across every signal. Equity exposure was at maximum. Mean reversion engines were quiet -- there was nothing stretched enough to snap back. The NLP filter was reading earnings transcripts from Q4 and finding the usual mix of confidence and deflection. Nothing unusual. Nothing alarming.

This is worth noting: the system did not predict COVID-19. It had no pandemic model, no epidemiological data feed, no early-warning satellite imagery of Wuhan hospitals. On February 19, the system was as bullish as everyone else.

The difference was not in what it knew. The difference was in what it was willing to let go of.

Five Days That Changed Everything

On Monday, February 24, the S&P 500 dropped 3.4%. By itself, that is a notable but unremarkable event -- the kind of pullback that happens several times a year. The system registered the move but did not change its regime classification. A single red day does not make a crisis.

But the system watches more than price. It watches the relationships between prices. And on February 24, something happened that price alone could not reveal: correlations began rising across asset classes. Stocks that normally moved independently started falling together. Sectors that typically offset each other -- technology and utilities, cyclicals and defensives -- were suddenly moving in lockstep.

By Wednesday, February 26, the VIX had doubled. The system's regime detector, which synthesizes ten independent signals including volatility, correlation structure, credit spreads, and momentum breadth, shifted its classification from "bull" to "stress." This was not a prediction. It was a measurement. The world had already changed; the system was simply the first to admit it.

The rotation began immediately. Equity exposure started dropping -- not to zero, but proportionally, engine by engine, based on each strategy's historical behavior in stress environments. Gold allocation rose. The mean reversion engine activated, preparing to capture the violent snaps that stretched prices inevitably produce. The CTA trend engine began reading the new momentum: downward.

By Friday, February 28, the S&P had fallen 12.8% in five trading days -- the fastest correction from an all-time high in history. The VIX had tripled from its February 19 reading. Most portfolio managers were still deciding whether this was a buying opportunity.

The system had already moved.

The Abyss

The first two weeks of March 2020 will be studied for generations. On March 9, the S&P 500 fell 7.6% in a single session -- triggering the first market-wide circuit breaker since 1997. Trading was halted for fifteen minutes. When it resumed, selling continued. Three days later, on March 12, the market fell another 9.5%. Another circuit breaker. Another halt.

The VIX hit 82.69 on March 16 -- a number so extreme that standard risk models literally could not process it. Many institutional risk systems have hard-coded assumptions that the VIX cannot exceed 80. On that Monday, those assumptions failed. The models broke. The humans behind them froze.

The system's regime detector had already shifted to full "crisis" classification. This matters because crisis is not simply a more intense version of stress. It is a qualitatively different market state with its own dynamics. In a crisis, the strategies that dominate are not the ones that work in bull markets or even in ordinary corrections. They are the ones designed to harvest energy from dislocation.

Mean reversion was the star. When the S&P 500 falls 9.5% in a day, individual stocks stretch far beyond any historical norm. Prices that should be at 80 trade at 50. Prices that should be at 200 trade at 120. The mean reversion engine does not need to know why the panic happened. It needs to know that prices are stretched -- and that stretched prices, across thousands of historical observations, snap back. During the worst two weeks of the crisis, the mean reversion engine returned +18%.

Crypto carry went flat. The digital asset markets experienced their own form of circuit breaker -- exchanges went down, liquidity vanished, and the carry engine correctly identified that the conditions for its strategy had evaporated. It parked capital and waited. This is worth noting: an engine that returns 0% during a crisis is not failing. It is surviving.

The CTA trend engine caught both sides of the volatility -- the momentum down as markets collapsed, and the first hints of reversal as the Fed intervened. Its net contribution was modest but positive: +2%.

The Bottom

The week of March 16-23 was the worst week for the S&P 500 since the 2008 financial crisis. By March 23, the index had fallen 34% from its all-time high just twenty-three trading days earlier. Trillions of dollars in wealth had evaporated. Retirement accounts were decimated. The 60/40 portfolio -- the bedrock of conventional financial planning -- was hemorrhaging on both sides as bonds, ordinarily the counterweight to falling stocks, offered almost no protection.

But the system was already positioned in the assets absorbing the energy. Gold, which had briefly dipped in the initial liquidity crunch, was rising. Treasury bonds at the long end were rallying as investors fled to safety. The volatility itself -- measured by VIX futures -- had become an asset class that the system was implicitly long through its regime-conditional allocations.

The equity strategies that remained active were positioned in the specific sectors showing resilience: technology companies enabling remote work, healthcare companies racing for treatments, consumer staples that people needed regardless of lockdowns. The NLP filter, which had scored Q4 2019 earnings calls, had flagged several airline and hospitality CEOs whose language patterns showed defensive rigidity -- exactly the characteristic that predicts poor performance under stress.

The net result: while the S&P 500 fell 34% peak to trough, the system returned +2.8% over the full drawdown period. Not because it predicted the virus. Not because it had insider knowledge. But because it runs eleven independent engines, each responding to its own signals, and the engines that thrive in crisis more than offset the engines that suffer.

Eleven Engines, One Crisis

The following chart shows the contribution of each engine during the COVID-19 drawdown period. Some engines lost money. That is the design. A system where every component wins in every environment is a system that has been overfit to history. The portfolio survives not because every engine works all the time, but because the right engines work at the right time.

Five engines gained. Four engines lost. Two were flat. The net was positive -- not because the system avoided loss, but because the gains from crisis-responsive engines overwhelmed the losses from equity-sensitive ones. This is what genuine diversification looks like. Not owning different colored versions of the same bet, but owning fundamentally different responses to the same reality.

The Deeper Question

There is a natural temptation to frame the COVID response as a success of speed -- the system moved faster than humans could. And that is partially true. By the time most portfolio managers had scheduled their first emergency meeting, the system had already rotated half its exposure. The latency between signal and action was measured in hours, not days.

But speed is the surface story. The deeper story is about attachment.

Every major portfolio blowup in history shares one characteristic. Not bad math. Not insufficient data. The inability to accept that reality had changed. LTCM's models were brilliant -- but they could not accept that Russian bonds could default. Tiger Management's analysis was correct -- but it could not accept that the market would stay irrational longer than its capital could stay solvent. Melvin Capital's thesis on GameStop was probably right on fundamentals -- but it could not accept that the game had changed.

The system has no thesis to defend. It has no ego invested in being right about the previous regime. When the VIX moved from 14 to 40 in five days, the system did not ask "why" or "is this rational" or "will it revert." It simply read the new state of the world and allocated accordingly. The pre-COVID world was gone. The system let it go instantly because it had never been holding on in the first place.

2022: When Both Pillars Fell

If the COVID crash was a speed test, 2022 was an endurance test. The Federal Reserve began raising interest rates in March 2022 to combat inflation that had reached 9.1% -- the highest in four decades. Over the course of the year, something happened that the conventional investment framework was not built to handle.

Stocks and bonds fell together.

The 60/40 portfolio -- 60% stocks, 40% bonds -- is the foundation of virtually every institutional investment program, every target-date retirement fund, every advisory practice in the developed world. Its premise is simple: when stocks fall, bonds rise, cushioning the blow. For forty years, that premise held. In 2022, it failed.

The S&P 500 fell 18.1%. Long-term Treasury bonds fell 31.2%. The 60/40 blend lost roughly 16%. Retirees who had been told their portfolios were "balanced" watched both halves decline simultaneously. The promise of diversification -- the entire intellectual framework of modern portfolio management -- appeared to break.

Prism returned +7.3%.

The reason is structural. Prism's eleven engines are not distributed across the stock-bond axis. They are distributed across independent sources of return -- mean reversion, momentum, carry, regime rotation, language analysis, trend following. When stocks and bonds fell together, the system's mean reversion engine was harvesting energy from dislocated prices. The regime rotation engine had shifted allocation toward assets benefiting from rising rates. The CTA trend engine was capturing the persistent downward momentum in bonds. The crypto carry engine was collecting funding premiums that had nothing to do with either stocks or bonds.

The engines that lost money in 2022 were different from the ones that lost money in 2020. And the engines that made money in 2022 were different from the ones that made money in 2020. That is the entire point. The system is not designed to have one answer for all markets. It is designed to always have some engines reading the current reality correctly -- wherever the energy is flowing.

The Lesson

There is a question that surfaces in every conversation with a prospective partner: "What happens in the next crisis?"

The honest answer is: we do not know what the next crisis will look like. It might be a pandemic. It might be a sovereign debt default. It might be an AI-related flash crash, or a geopolitical shock, or a monetary system event that nobody has modeled. The nature of genuine black swans is that they are, by definition, unpredictable.

But the system does not need to predict the nature of the crisis. It needs to do two things: measure the new reality quickly and have engines that profit from dislocation. The measurement infrastructure -- ten independent regime signals updating daily -- gives it speed. The engine diversity -- eleven strategies with an average pairwise correlation near zero -- gives it optionality.

The next crisis will be different from COVID and different from 2022. The system's response will also be different. Some engines will contribute that were quiet before. Others will lose money that gained last time. The specific engines that dominate are unpredictable.

The net result is not.

In the two worst market environments of the past two decades -- environments that were completely different in character, cause, and duration -- the system was positive. Not because it predicted either one. Because it had nothing to let go of, and everything to read.