The March 2020 Anomaly
On March 23, 2020, the S&P 500 hit bottom at 2,237 after a historic 34% freefall from its February 19 peak of 3,386. The fastest bear market in history. Hedge funds across strategies hemorrhaged capital — equity long/short funds were down sharply, the HFRX Global Hedge Fund Index fell 6.90% year-to-date through March, and panic selling dominated every asset class.
Except one corner of the market was making money.
Commodity Trading Advisors (CTAs) and managed futures funds — strategies built on Markov Decision Processes and regime-switching models — were posting gains. The Barclay CTA Index returned +1.94% in March 2020 while most strategies collapsed. The HFRX Macro Systematic Diversified CTA Index posted +1.11% during the same period. These weren’t small profits from luck. This was systematic alpha extraction during maximum chaos, a phenomenon the industry calls “crisis alpha.” By year’s end, while the average hedge fund returned 11.14%, many leading CTAs and volatility strategies delivered strong double-digit returns.
The question isn’t whether they made money. The question is how — and specifically, how the mathematics of Markov processes enabled it.
Foundation: Understanding Markov Processes
Before explaining how CTAs printed money in March 2020, we need to understand the theoretical engine: the Markov process.
The Core Property
A Markov process is a stochastic system where the future depends only on the present state, not the path taken to get there. Mathematically, for a random process {Xₙ, n = 0, 1, 2, …}:
P(Xₘ₊₁ = j | Xₘ = i, Xₘ₋₁ = iₘ₋₁, …, X₀ = i₀) = P(Xₘ₊₁ = j | Xₘ = i)
This “memorylessness” is the Markov property: only the current state matters for predicting the next state.
A Practical Example: Stock Price Evolution
Consider a simplified stock price model where from time t to t+1, the price can either:
Go up by 1: Xₜ₊₁ = Xₜ + 1
Go down by 1: Xₜ₊₁ = Xₜ − 1
To model this as a Markov process, we define the probability of an up-move as a function of the current state. One elegant approach uses a logistic function:
P[Xₜ₊₁ = Xₜ + 1] = 1 / (1 + e^(-α₁(L — Xₜ)))
Where:
L is a reference level (e.g., long-term mean)
α₁ is a “pull strength” parameter controlling mean reversion
This creates mean-reverting behavior: when the price Xₜ is far below L, the probability of an up-move increases. When it’s far above L, the probability of an up-move decreases. The process “remembers” nothing except where it is right now.
The Implementation
Simulating this process is straightforward:
def prob_function(alpha):
return lambda x: 1 / (1 + np.exp(-alpha * x))
def next_price(current_price, level_param, alpha):
logistic = prob_function(alpha)
up_prob = logistic(level_param - current_price)
up_move = np.random.binomial(1, up_prob)
return current_price + (2 * up_move - 1) At each time step, we sample from a Bernoulli distribution with success probability equal to the up-move probability. If we get 1, the price increases by 1; if 0, it decreases by 1.
This toy model captures a critical insight: markets exist in states, and transition probabilities between states are predictable.
From Theory to Reality: Regime-Switching Models
The simple Markov model above assumes a single “regime” with constant parameters. Real markets don’t work this way. Volatility spikes during crises. Correlations break down. Trends emerge and disappear. Markets shift between fundamentally different regimes.
This is where Hidden Markov Models (HMMs) and Markov Regime-Switching Models become powerful.
Hidden Markov Models in Finance
In an HMM, there are:
Observable states: Asset prices, returns, volatility
Hidden states: Underlying market regimes (bull market, bear market, high volatility, low volatility)
Transition probabilities: Likelihood of switching from one regime to another
The “hidden” regimes drive the observed price behavior, but you can’t directly observe which regime you’re in — you must infer it from price data.
Academic research shows HMMs can effectively identify distinct market regimes:
Research from Cambridge University (2020) demonstrated that Hidden Markov Models applied to intraday momentum trading can identify regime shifts and generate statistically significant returns with Sharpe ratios of 1.9, specifically because they avoid the time-lagging problem of traditional filters
A 2019 study on S&P 500 high-frequency returns found that regime-switching models in pairs trading achieved an annualized Sharpe ratio of 3.92 after transaction costs by automatically detecting regime changes
A 2015 analysis showed that Markov regime-switching models in pairs trading outperformed traditional distance methods, especially during the 2008 financial crisis
The key advantage: regime-switching models detect when the market’s behavior has fundamentally changed — and adjust positions accordingly.
How CTAs Used Markov Models to Profit in March 2020
Now we arrive at the core question: How did this theoretical framework translate into real profits during the COVID crash?
The CTA Strategy: Trend Following with Regime Detection
CTAs (Commodity Trading Advisors) manage approximately $350–400 billion in assets through managed futures strategies (with industry estimates varying by source and year). The dominant approach is trend following — going long assets in uptrends and short assets in downtrends.
But not just any trend following. Sophisticated CTAs use multi-regime trend-following systems that adjust based on detected market states:
Normal regime: Trade established trends with moderate position sizing
Crisis regime: Rapidly de-risk losing positions, aggressively size winning positions, reduce exposure to correlated markets
The March 2020 Trade Anatomy
As COVID-19 paralyzed the global economy in February-March 2020, regime-switching models in CTA systems detected a fundamental state change. Here’s what happened:
Phase 1: Regime Detection (Late February 2020)
Volatility spiked dramatically. The VIX jumped from the low teens to over 40 within approximately two weeks, eventually peaking at 82.69 on March 16. Correlations across equity markets surged toward 1.0. Regime-switching models flagged a transition from “normal volatility” to “crisis volatility.”
Research on CTA performance during COVID confirms this: “COVID-19 did not only drive trends; it also (temporarily) changed market dynamics. Dealing with changed market dynamics was probably the biggest challenge… CTAs were able to rapidly rebalance their portfolios in response to significant market and performance changes.”
Phase 2: Rapid Position Adjustment (Early March 2020)
The Markov framework’s strength emerged: fast adaptation to regime shifts.
According to a 2022 study on CTA crisis alpha: “The fast reduction in CTAs’ exposure to crisis markets (in less than 15 days for composite indices) allows them to stabilize their performance.”
CTAs executed three simultaneous moves:
A) Flipped equity positions from long to short
Trend-following signals turned decisively negative
Systems went from net long equities to net short within days
As stocks crashed, short positions generated profits
B) Went long safe-haven assets
Increased exposure to U.S. Treasuries (bonds rallied as yields collapsed)
Went long gold (up 25% from March lows to August highs)
Captured the “flight to safety” trend
C) Shorted commodities
Oil crashed from $60/barrel to negative $37 (unprecedented)
CTAs were massively short crude oil futures
Captured one of the fastest commodity crashes in history
Phase 3: Diversification Across Uncorrelated Markets (Mid-March 2020)
This is where the Markov framework’s multi-market structure created alpha. Research on CTA crisis performance states: “Positive yields in gaining markets can counterbalance low performance in the crisis market. Both factors together, quickly cutting losses in crisis sectors while staying profitable in the other ones, allow CTAs to generate positive crisis returns.”
While equity-focused hedge funds were trapped in correlated losses (everything down), CTAs operated across:
20+ global equity indices
10+ bond markets
15+ currency pairs
25+ commodity futures
The regime-switching models independently assessed each market’s state. When U.S. equities were in “crash regime,” German bunds were in “rally regime.” When oil was in “collapse regime,” gold was in “flight-to-safety regime.”
The P&L Mechanics: Where the Money Came From
Let’s quantify how the strategy generated returns:
Position 1: Short S&P 500 E-mini Futures
Peak: ~3,386 (February 19, 2020)
Entry: ~3,230 (late February 2020) as downtrend confirmed
Bottom: ~2,237 (March 23, 2020)
Exit: ~2,480 (late March 2020)
Profit: ~23% on notional from entry to exit (amplified by leverage)
With 3:1 leverage: ~69% return on capital allocated to this position
Position 2: Long U.S. 10-Year Treasury Futures
Entry: Yield ~1.5% (late February 2020)
Peak: Yield hit historic low of 0.318% (March 9, 2020)
Exit point: Yield ~0.5–0.7% (mid-to-late March 2020)
Bond prices rally when yields fall
Profit: ~8–10% on notional, amplified by leverage
Position 3: Short Crude Oil Futures
February 2020: Oil trading around $60/barrel
Entry: ~$53/barrel (early March 2020) as initial collapse began
Exit: ~$20/barrel (late March 2020)
Profit: ~62% on notional position
Note: Oil continued crashing to historic negative -$37/barrel on April 20, 2020
Position 4: Long Gold Futures
Entry: ~$1,580–1,586/oz (early March 2020)
Exit: ~$1,680/oz (late March 2020)
Profit: ~6% on notional (maintained position for larger gains later)
Note: Gold continued rising to over $2,000/oz by August 2020
Aggregate Result: By maintaining a diversified book across these positions with appropriate risk management, CTAs generated positive returns while most strategies collapsed.
Data confirms this: “During the March 2020 crash, managed futures funds were mostly long U.S. Treasuries and gold, and short crude oil, offsetting the negative impact of their long-time bias in equities.”
The Statistical Evidence: Crisis Alpha is Real
The performance data from March 2020 validates the Markov regime-switching approach:
Aggregate Industry Performance
CTA/Managed Futures: Positive returns during March 2020 crash
Equity Long/Short Funds: Down sharply in March, with the HFRI Equity Hedge Index falling approximately 9.5% for the month
HFRX Global Hedge Fund Index: Down 5.88% in March, -6.90% YTD through March
S&P 500: Down 12.51% in March, -20% YTD through March 23
Specific Strategy Performance
Research on hedge fund strategies during COVID shows: “Equity long bias, volatility trading, convertible arbitrage, and emerging market funds delivered gains in excess of 15%” in 2020, with managed futures among the top performers.
A 2022 study found: “CTAs do acquire positive gains in most sectors during crises, which originate from two sources: Firstly, their diversification across multiple futures markets… Secondly, the fast reduction in CTAs’ exposure to crisis markets.”
The Recovery Phase (Q2-Q4 2020)
The Markov framework’s adaptability shone through as regimes shifted again:
Q2 2020: Markets rebounded on Fed intervention. Regime models detected the shift from “crisis” to “recovery”
CTAs flipped from net short to net long equities
Captured the V-shaped recovery rally
By Q3 2020, the hedge fund industry “completely offset the losses they had incurred due to the COVID-19 crisis”
Q4 2020: Industry grew another 13.22%
The regime-switching framework allowed CTAs to profit in both directions: down during the crash, up during the recovery.
Why Traditional Models Failed Where Markov Models Succeeded
The stark performance divergence in March 2020 reveals why state-dependent models matter:
Traditional Static Models
Most hedge funds operated with fixed assumptions:
Constant correlation matrices
Static risk parameters
Single-regime optimization
When COVID hit, these assumptions shattered:
Correlations that were 0.3 in January became 0.9 in March
Volatility that was 15% annualized became 80%
Market liquidity that was abundant vanished overnight
Markov Regime-Switching Advantage
The regime-switching framework succeeded because it expected regime changes:
No Fixed Parameters: Volatility, correlation, and trend speeds adjusted dynamically based on detected regime
Explicit State Modeling: Systems explicitly modeled “crisis states” separate from “normal states”
Rapid Adaptation: Transition probabilities allowed sub-15-day position adjustments
Research on hedge fund timing during COVID confirms: “By assuming imperfect information about financial market states captured by a two-state Markov regime switching process, we introduce a learning process into hedge fund managers’ risk choice problem.”
This isn’t just academic theory — it’s the operational framework that separated winners from losers in March 2020.
Lessons for Quantitative Traders
The March 2020 case study offers several actionable insights:
Lesson 1: Model Regime Shifts, Don’t Assume Stationarity
The Failure Mode: Assuming market parameters stay constant
The Solution: Implement regime-switching models that explicitly allow for state transitions
Traditional GARCH models assume volatility clustering but not fundamental regime changes. Hidden Markov Models and Markov regime-switching models go further — they assume the entire data-generating process can shift.
Implementation: Use HMMs to classify market states (high/low volatility, trending/mean-reverting, crisis/normal), then optimize strategy parameters separately for each regime.
Lesson 2: Speed Matters in Crisis Detection
The Evidence: CTAs reduced crisis market exposure in under 15 days
The Implication: Your regime detection must work on short time horizons (days to weeks, not months)
Research on COVID-era CTA performance found: “Although overall performance may have been better employing longer-term models, in years like 2020, during the COVID crisis, the faster models delivered the best hypothetical returns.”
Implementation: Use multiple time-horizon models (fast: 20–60 day MAs, slow: 100–300 day MAs). In detected crisis regimes, weight fast models higher.
Lesson 3: Diversification Across Uncorrelated Markets Is Non-Negotiable
The Data: CTAs profited because they traded 50+ uncorrelated markets simultaneously
The Failure Mode: Being trapped in a single asset class during regime-wide correlation spikes
During March 2020, all equities moved together (correlation → 1.0). But gold, treasuries, and currencies maintained distinct behaviors. Only multi-market strategies captured this.
Implementation: Build strategies that operate across:
Multiple equity markets (not just S&P 500)
Fixed income (government and corporate)
Currencies (G10 and emerging market)
Commodities (energy, metals, agriculture)
Lesson 4: Crisis Alpha Comes From Asymmetric Regime Behavior
The Insight: Markets crash faster than they rally. Regime models capture this asymmetry.
Research shows: “Trend-following managers still maintain faster models in their portfolios to better handle periods of sudden market stress and reversals.”
The March 2020 drawdown took 33 days (Feb 19 — Mar 23). The recovery took 5 months. Regime-switching models captured both moves because they detected the regime change — not because they predicted the duration.
Implementation:
In high-volatility regimes: Tighten stop losses, reduce position sizes, increase monitoring frequency
In low-volatility regimes: Widen stop losses, increase position sizes, reduce monitoring frequency
Lesson 5: The “Crisis Alpha” Label is Regime Beta
An important theoretical point: What the industry calls “crisis alpha” is actually regime-specific beta.
Research on Transtrend’s COVID performance stated: “The ‘crisis alpha’ inherent to trend following CTA programs should be decomposed into a ‘crisis beta’ that comes with the investment style, and a potentially significant amount of negative ‘crisis alpha’ that explains the typically huge dispersion between correlated CTA programs during crisis periods.”
This matters for strategy design: You’re not trying to predict crises. You’re trying to detect regime transitions and capture regime-specific beta. The March 2020 profits came from:
Correctly identifying the regime shift
Quickly repositioning to capture regime-specific trends
Avoiding regime-blind strategies
Building Your Own Regime-Switching Strategy: A Practical Framework
Here’s how to implement the lessons from March 2020:
Step 1: Define Your Regime States
Start with a 2-state or 3-state model:
2-State Model:
State 1: Normal volatility (VIX < 20)
State 2: Crisis volatility (VIX > 30)
3-State Model:
State 1: Low volatility, mean-reverting (VIX < 15)
State 2: Normal volatility, trending (VIX 15–25)
State 3: High volatility, crisis (VIX > 25)
Step 2: Estimate Transition Probabilities
Use historical data to estimate P(State i → State j). For example:
P(Normal → Crisis) = 5% (crisis regimes are rare)
P(Crisis → Normal) = 15% (crises resolve faster)
P(Normal → Normal) = 95% (persistence)
Train an HMM on historical data using the Baum-Welch algorithm (available in Python libraries like hmmlearn).
Step 3: Regime-Conditional Trading Rules
Define different strategy parameters for each regime:
Normal Regime:
Position size: 100% of standard
Stop loss: 2 standard deviations
Lookback period: 100 days
Crisis Regime:
Position size: 50% of standard (reduce risk)
Stop loss: 1 standard deviation (tighter risk control)
Lookback period: 20 days (faster adaptation)
Step 4: Backtesting Framework
Test your regime-switching strategy against:
March 2020 COVID crash
February 2018 “Volmageddon”
August 2015 China devaluation
October 2008 Lehman collapse
If your regime detection successfully identified these shifts and adjusted positions, you’ve replicated the framework that generated March 2020’s crisis alpha.
Step 5: Implementation Considerations
Data Requirements:
High-frequency regime indicators (daily at minimum)
Multiple asset classes for diversification
Transaction cost modeling (slippage in crisis regimes increases)
Risk Management:
Maximum drawdown limits per regime
Leverage constraints (reduce leverage in detected crisis regimes)
Liquidity monitoring (some markets become illiquid during regime shifts)
The Broader Context: Why Regime-Switching Models Matter Now
The March 2020 case study isn’t just historical trivia. It’s a template for the current market environment.
Post-2020: A Higher-Volatility Regime
Research on hedge funds in the new macro regime states: “A new market regime characterized by higher volatility, interest rates, and inflation has emerged since 2022… hedge funds have so far boasted a 1.34 Sharpe ratio compared to 0.77 for the 60/40 portfolio.”
We’ve shifted from:
2010–2020: Low volatility, persistent trends, ZIRP (Zero Interest Rate Policy)
2022-present: Higher volatility, rapid regime changes, inflation concerns
In the first era, static models worked fine because regimes rarely changed. In the current era, regime-switching models are essential because markets oscillate between distinctly different states.
The Dispersion Opportunity
Research on CTA return dispersion in 2024–2025 found: “The range of returns [between CTAs] was nearly 15%, and has been consistently high each year.”
This dispersion reflects implementation quality. Funds using sophisticated regime-switching models outperformed by wide margins. The math matters.
Conclusion: From Theory to Practice
The journey from the simple Markov process in the introduction to the complex regime-switching models that profited in March 2020 reveals a critical insight about quantitative finance:
Markets are state-dependent systems. Strategies that explicitly model and adapt to state changes outperform strategies that assume stationarity.
The CTAs that made money during COVID’s crash weren’t lucky. They were using mathematical frameworks — Hidden Markov Models, Markov regime-switching processes, multi-state transition matrices — that explicitly modeled the possibility of regime changes.
When the regime changed, their models detected it. When the models detected it, their systems adapted. When their systems adapted, they repositioned from long equities to short equities, from short bonds to long bonds, from neutral commodities to short oil/long gold.
The result: positive returns during one of the fastest market crashes in history.
This is what Markov processes buy you in real trading: the ability to profit from state transitions that break everyone else’s assumptions.
The math works. The only question is whether you’re using it.
References & Further Reading
Academic Research Cited
Hidden Markov Models Applied To Intraday Momentum Trading (Christensen, Turner, Godsill, 2020) — arXiv:2006.08307
Hedge Fund Treasury Trading and Funding Fragility: Evidence from the COVID-19 Crisis (Federal Reserve, 2021)
A flexible regime switching model with pairs trading application to the S&P 500 high-frequency stock returns (Endres & Stübinger, 2019)
The crisis alpha of managed futures: Myth or reality? (ScienceDirect, 2022)
American hedge funds industry, market timing and COVID-19 crisis (Journal of Asset Management, 2022)
Pairs trading: The performance of a stochastic spread model with regime switching (Yang et al., 2016)
Industry Data Sources
HFR Indices — Hedge Fund Research performance data
HFRX Global Hedge Fund Index March 2020 Performance Notes
HFRI Equity Hedge Index March 2020 Performance Notes
2. BarclayHedge — Managed futures industry benchmarks
Barclay CTA Index March 2020 returns (+1.94%)
BTOP50 Index (largest 50% of investable CTA assets)
3. SG CTA Trend Index — Société Générale CTA performance tracking
4. MSCI Hedge Fund Intel — Hedge fund positioning data during COVID-19
5. Traders Magazine — “Barclay CTA Index Returns 1.94% in March” (April 2020)
6. Institutional Investor — “Proven Hedge Funds Excelled During the Pandemic” (Average hedge fund 2020 return: 11.14%)
Practical Implementation
QuantConnect — Research on HMM intraday trading (2022–2023 backtests)
Transtrend — “Crisis alpha in the Covid-19 crisis” (practitioner perspective)
CFM — “Steady Trends: The Reality of CTA Return Dispersion” (2025)
About This Analysis: This article synthesizes academic research, industry data, and documented hedge fund performance to explain how Markov-based regime-switching models enabled specific trading strategies to profit during the March 2020 market crash. All performance claims are supported by cited sources. The mathematical framework presented (Markov processes, HMMs, regime-switching models) represents the actual methodology used by quantitative hedge funds managing hundreds of billions in managed futures strategies.
Data Verification Notes:
March 2020 performance data verified through HFR official performance notes and BarclayHedge indices
Barclay CTA Index: +1.94% in March 2020, +1.88% YTD (Source: Traders Magazine)
HFRX Global Hedge Fund Index: -5.88% in March 2020, -6.90% YTD through March (Source: HFR Performance Notes)
HFRI Equity Hedge: -9.5% to -9.58% in March 2020 (Source: HFR Performance Notes)
S&P 500: Peak of 3,386 on February 19, bottom of 2,237 on March 23 (34% decline); -12.51% for March month (Source: Bankrate, Forbes, Managed Futures Investing)
VIX: Peaked at 82.69 on March 16, 2020 (Source: Macroption, CNBC)
Oil prices: ~$60/barrel in February, ~$53 in early March, ~$20 by late March, crashed to negative -$37/barrel on April 20, 2020 (WTI futures) (Source: Oil Price, Investopedia, Facebook/Daily Star Zambia)
Gold: ~$1,580–1,586/oz in early March 2020 (Source: Bullion-Rates, Statmuse)
10-Year Treasury yield: Historic low of 0.318% on March 9, 2020 (Source: CNBC)
Average hedge fund 2020 return of 11.14% verified through BarclayHedge data (Source: Reuters, Institutional Investor)
The content is for educational purposes only and does not constitute investment advice. Past performance does not guarantee future results.
Cover photograph: Håkan Dahlström from Malmö, Sweden, CC BY 2.0, via Wikimedia Commons.