Home AI Trading Strategies / Mean-Reversion Clusters

Mean-Reversion Clusters Trading Strategy Analysis

Identify and analyze price clusters with Sourcetable AI. Calculate reversion probabilities, optimize entry points, and backtest strategies automatically using natural language.

Andrew Grosser

Andrew Grosser

February 24, 2026 • 17 min read

Introduction

October 2022: SPY falls to $352. It crashed below its 200-day MA, reached 2 standard deviations below the 20-day mean, and RSI hit 26. Three separate mean-reversion signals clustering simultaneously. You've noticed something interesting in your stock data. Prices don't move randomly—they cluster around certain levels, forming support and resistance zones that traders watch obsessively. When prices deviate significantly from these clusters, they tend to snap back like a rubber band. This is mean-reversion cluster trading, a quantitative strategy that profits from identifying price concentrations and betting on returns to these levels.

Traditional cluster analysis requires complex statistical software, custom scripts, and hours of manual calculation. You need to identify density peaks in price distributions, calculate standard deviations from cluster centers, determine optimal entry thresholds, and backtest thousands of scenarios. Most traders rely on Excel with dozens of formulas for clustering algorithms, z-score calculations, and probability distributions. One wrong cell reference and your entire analysis collapses sign up free.

Why Sourcetable for Mean-Reversion Cluster Analysis

Mean-reversion cluster trading identifies price levels where significant trading activity has concentrated, then profits when prices deviate from and return to these levels. The strategy assumes that prices gravitate toward equilibrium zones where buyers and sellers have historically agreed on value. When external shocks push prices away from these clusters, informed traders recognize the dislocation and position for the inevitable return.

The mathematical foundation involves kernel density estimation to identify price concentrations, statistical distance measurements to quantify deviations, and probability calculations to assess reversion likelihood. A typical setup: TSLA trades in a cluster around $245-$255 for three months with 60% of volume occurring in this range. When news pushes the stock to $275, that's 2.1 standard deviations from the cluster center. Historical data shows 78% reversion probability within 10 trading days. You short at $275 with a target of $255 and stop-loss at $285.

Excel makes this analysis painful. You need FREQUENCY functions for distribution analysis, complex array formulas for kernel density estimation, STDEV.P calculations for deviation metrics, and VLOOKUP chains for historical reversion rates. Want to test different cluster timeframes? Rebuild everything. Need to analyze 50 stocks simultaneously? Copy formulas across massive worksheets and pray nothing breaks. Backtesting requires VBA macros or external tools.

Sourcetable eliminates every friction point. The AI understands clustering concepts natively—upload your price data and ask 'Identify price clusters for the past 6 months using 95% confidence intervals.' It automatically performs density analysis, identifies statistically significant clusters, calculates center points and boundaries, and visualizes results with color-coded heatmaps. Ask 'What's the current deviation from the primary cluster?' and get instant z-scores with historical reversion probabilities.

The platform handles multi-asset analysis effortlessly. Upload data for your entire watchlist and ask 'Which stocks are currently 2+ standard deviations from their main cluster?' Sourcetable scans everything, ranks opportunities by deviation magnitude and reversion probability, and presents actionable results in seconds. Want to backtest your entry rules? Ask 'Show me returns from entering at 1.5 SD and exiting at cluster center over the past year.' The AI runs the simulation and shows win rate, average return, maximum drawdown, and Sharpe ratio.

Real-time monitoring becomes simple. Connect live price feeds and ask 'Alert me when any position deviates beyond 2 SD from its cluster.' Sourcetable watches your portfolio continuously and notifies you of opportunities. The AI even helps with position sizing: 'How much should I allocate to this trade given 72% reversion probability and $500 risk budget?' It calculates optimal position size using Kelly criterion or fixed fractional methods.

Benefits of Cluster Analysis with Sourcetable

Mean-reversion cluster strategies offer systematic, data-driven trading with quantifiable edge. Instead of guessing support and resistance levels, you're identifying statistically significant price zones backed by volume and time data. Organizations from proprietary trading firms to hedge funds use cluster analysis to find high-probability setups with favorable risk-reward ratios.

Automated Cluster Identification

Sourcetable's AI automatically detects price clusters using sophisticated density algorithms without requiring statistical expertise. Upload historical prices and ask 'Find all significant clusters from the past year.' The system performs kernel density estimation, identifies local maxima in the distribution, applies statistical significance tests, and returns cluster centers with confidence intervals. You get results like 'Primary cluster: $127.50-$132.50 (68% of volume), Secondary cluster: $118-$122 (22% of volume)' with visual distribution charts showing density peaks.

The AI adapts to different market conditions automatically. In trending markets, it identifies fewer, wider clusters. In range-bound periods, it detects tight, well-defined clusters with higher reversion probabilities. Ask 'How has cluster structure changed over time?' and see temporal evolution showing when markets transitioned from trending to mean-reverting behavior. This context helps you adjust strategy parameters for current conditions.

  • K-Means Clustering: Groups price states into K clusters based on multiple features (return, volume, RSI, spread vs MA); points within the same cluster behave similarly historically - when current conditions match a historically mean-reverting cluster, enter the trade.
  • Feature Space: Typical features for mean-reversion clustering: 5-day return, 20-day return, RSI(14), Bollinger Band position, volume ratio; normalize all features to [0,1] before clustering to prevent scale dominance.
  • Cluster Labeling: After clustering, label each cluster by its average forward 5-day and 20-day returns; clusters with average forward 5-day return above +1.5% and Sharpe above 1.0 are buy clusters; clusters with average forward return below -1.5% are sell clusters.
  • Silhouette Score: Measures cluster quality; ranges from -1 to +1; above 0.5 = good cluster separation; below 0.3 = overlapping clusters with weak predictive signal; accept only clusters with silhouette above 0.4 for trading signals.

Real-Time Deviation Monitoring

Track price deviations from cluster centers continuously without manual calculations. Sourcetable calculates z-scores in real-time, showing exactly how far current price has moved from equilibrium. Ask 'What's the current deviation status for my watchlist?' and get instant results: 'AAPL: +1.8 SD (approaching entry threshold), MSFT: -0.4 SD (neutral), GOOGL: +2.3 SD (active signal).' The AI highlights which positions meet your entry criteria and estimates reversion probability based on historical patterns.

Set custom alert thresholds that match your risk tolerance. Tell Sourcetable 'Notify me when any stock moves beyond 1.5 standard deviations from its primary cluster' and the system monitors everything automatically. When triggers hit, you receive alerts with full context: deviation magnitude, historical reversion rate at this level, average time to reversion, and suggested position sizing. This transforms cluster trading from reactive to proactive—you catch opportunities the moment they appear.

  • Z-Score Threshold: Current price minus rolling mean divided by rolling standard deviation; z-score below -2.0 on a 20-day window historically produces positive forward returns 67% of the time for large-cap US stocks (1990-2023 backtest).
  • Multi-Indicator Convergence: When 3+ mean-reversion indicators simultaneously signal (RSI below 30 AND z-score below -2 AND below Bollinger lower band AND above-average volume), win rate increases from 67% to 78% - signal convergence is the edge.
  • Deviation Half-Life: Estimate how long the deviation persists before reverting; stocks with high autocorrelation (slow mean reversion) have longer holding periods; Ornstein-Uhlenbeck half-life formula: T_half = ln(2)/kappa where kappa is mean-reversion speed.
  • Adaptive Thresholds: Use rolling percentile-based z-scores rather than fixed plus/minus 2 standard deviation thresholds; during high-vol regimes, fixed thresholds trigger less often; adaptive thresholds maintain consistent signal frequency across volatility regimes.

Statistical Confidence Metrics

Every trade recommendation comes with quantified probability and confidence intervals. Ask 'What's the reversion probability for TSLA at current levels?' and Sourcetable analyzes historical instances where price deviated this far from the cluster. You get specific answers: '73% probability of returning to cluster center within 8 trading days (95% CI: 68-78%), average reversion move of $18.50, maximum adverse excursion of $6.20.' These metrics let you size positions appropriately and set realistic profit targets.

The AI distinguishes between high-confidence and low-confidence setups automatically. Clusters with consistent reversion patterns get higher probability scores. Clusters showing structural breakdown get flagged as unreliable. Ask 'Which opportunities have the strongest statistical support?' and Sourcetable ranks your watchlist by confidence score, filtering out marginal setups and highlighting only the highest-probability trades. This systematic approach removes emotional decision-making and keeps you focused on edge.

  • Augmented Dickey-Fuller Test: Tests whether a price series is stationary (mean-reverting) or random walk; ADF statistic below -3.0 (p-value below 0.05) confirms stationarity with 95% confidence - required before deploying mean-reversion strategies.
  • Half-Life Estimation: Regress daily price change on previous day price; slope coefficient beta = -kappa (mean-reversion speed); for SPY in 2022, beta=-0.08 implies half-life of ln(2)/0.08 = 8.7 trading days - optimal holding period for mean-reversion trades.
  • Hurst Exponent: H below 0.5 = mean-reverting, H=0.5 = random walk, H above 0.5 = trending; calculate via R/S analysis or variance ratio test; SPY daily returns show H around 0.48 (slightly mean-reverting); individual stocks range from H=0.40 to H=0.55.
  • Sample Size Requirements: Cluster analysis needs minimum 200+ observations in each cluster for statistically reliable forward return estimates; with 10 clusters over 5 years of daily data = 250 bars each, estimates have plus/minus 2% standard error - acceptable for strategy calibration.

Multi-Timeframe Cluster Analysis

Analyze cluster structure across different timeframes to identify nested opportunities. Ask 'Show me daily and weekly clusters for SPY' and Sourcetable reveals how short-term clusters exist within longer-term equilibrium zones. You might discover the daily cluster at $445-$450 sits at the lower boundary of the weekly cluster at $445-$465. This multi-timeframe confluence suggests strong support and higher reversion probability for long positions.

The platform makes timeframe comparison effortless. Request 'Compare 30-day vs 90-day cluster structures for my portfolio' and see side-by-side analysis showing which assets have aligned clusters (stronger signals) versus conflicting clusters (avoid or trade carefully). This context prevents you from fighting longer-term trends—you only take mean-reversion trades when clusters align across timeframes, dramatically improving win rates.

Automated Backtesting and Optimization

Test your cluster strategy rules across years of data with simple questions. Ask 'Backtest entering at 2 SD with targets at cluster center for AAPL since 2020' and Sourcetable simulates every trade, calculating win rate, average return per trade, maximum drawdown, Sharpe ratio, and profit factor. You see exactly how your rules would have performed through different market regimes—trending periods, volatile crashes, and range-bound consolidations.

Optimization happens through conversation. Ask 'What entry threshold produces the best risk-adjusted returns?' and the AI tests multiple deviation levels (1.0 SD, 1.5 SD, 2.0 SD, 2.5 SD), comparing performance metrics for each. You discover that 1.8 SD provides the optimal balance—frequent enough opportunities with strong enough edge. The system even warns about overfitting: 'Results are robust across different test periods' or 'Performance degrades significantly in out-of-sample data—consider different parameters.'

How Mean-Reversion Cluster Trading Works in Sourcetable

Implementing cluster strategies in Sourcetable follows a systematic workflow from data import through position management. The AI handles complex statistical calculations while you focus on strategy decisions and risk management. Here's the complete process for identifying and trading mean-reversion opportunities.

Step 1: Import and Prepare Price Data

Start by uploading historical price data for your target assets. Sourcetable accepts CSV files, Excel spreadsheets, or direct connections to data providers like Yahoo Finance, Alpha Vantage, or your broker's API. Your data should include date, open, high, low, close, and volume for each trading period. For cluster analysis, you typically need at least 3-6 months of daily data, though the AI adapts to whatever timeframe you provide.

Once uploaded, ask 'Show me the price distribution for NVDA over the past 6 months.' Sourcetable immediately creates a histogram showing frequency of prices at different levels. You'll see clear peaks where prices spent significant time—these are your cluster candidates. The AI overlays volume data to show which clusters had the most trading activity, indicating stronger support/resistance levels. This visual analysis takes seconds versus the hours required to build equivalent Excel charts.

  • Start by uploading historical price data for your target assets.
  • "Show me the price distribution for NVDA over the past 6 months."

Step 2: Identify Statistical Clusters

Ask Sourcetable to perform formal cluster identification: 'Identify significant price clusters using 95% confidence level.' The AI applies kernel density estimation to smooth the price distribution, identifies local maxima that represent cluster centers, calculates statistical significance using density thresholds, and returns results with precise boundaries. You get output like: 'Primary cluster: $485-$495 (center: $490, 58% of observations, p<0.01), Secondary cluster: $470-$478 (center: $474, 28% of observations, p<0.05).'

The system provides context for each cluster automatically. Ask 'Why is the $490 cluster significant?' and receive explanations: 'This cluster formed over 47 trading days with average daily volume 15% above baseline. Price tested this level 8 times with 6 successful reversions. Volume profile shows balanced buying/selling, indicating equilibrium zone.' This narrative helps you understand not just where clusters exist but why they're likely to persist.

Step 3: Calculate Deviation Metrics

With clusters identified, monitor current price position relative to cluster centers. Ask 'What's the current z-score for NVDA relative to its primary cluster?' Sourcetable calculates the standardized distance: 'Current price $510 is +2.1 standard deviations from primary cluster center at $490 (SD = $9.52).' The AI shows this graphically with color-coded zones: green for neutral (within 1 SD), yellow for moderate deviation (1-2 SD), and red for extreme deviation (2+ SD) where reversion trades become attractive.

For portfolio-wide monitoring, ask 'Show me all stocks with deviations exceeding 1.5 SD.' Sourcetable scans your entire watchlist, calculates z-scores for each asset, and returns a ranked table showing the most extreme deviations first. Each entry includes current price, cluster center, deviation magnitude, direction (above/below cluster), and time since deviation began. This systematic screening ensures you never miss high-probability opportunities across your universe.

  • "s the current z-score for NVDA relative to its primary cluster?"
  • "Show me all stocks with deviations exceeding 1.5 SD."

Step 4: Assess Reversion Probability

Before entering trades, quantify reversion likelihood using historical patterns. Ask 'What's the historical reversion rate when NVDA moves 2+ SD from its cluster?' Sourcetable analyzes all past instances matching current conditions, calculating: 'In 23 historical occurrences of 2+ SD deviations, price returned to within 0.5 SD of cluster center in 19 cases (83% success rate). Average time to reversion: 6.2 trading days. Average reversion move: $17.30. Maximum adverse excursion before reversion: $8.50.'

The AI adjusts probabilities based on context. If the broader market is trending strongly, reversion rates may be lower. If volatility is elevated, reversions may be faster but less reliable. Ask 'How do current market conditions affect reversion probability?' and get adjusted estimates: 'VIX at 28 (elevated) historically reduces reversion probability to 71% but increases average reversion magnitude to $21.50. Consider wider stops and larger profit targets.'

Step 5: Size Positions and Set Risk Parameters

Sourcetable helps you determine appropriate position size based on probability and risk tolerance. Tell the AI your constraints: 'I have $50,000 capital and want to risk no more than 2% per trade. What position size for NVDA short at $510 with stop at $520?' It calculates: 'Risk per share: $10. Maximum risk: $1,000 (2% of $50,000). Position size: 100 shares. Notional exposure: $51,000 (102% of capital—consider reducing to 50 shares for 1% risk if using margin).'

For more sophisticated sizing, ask 'Calculate optimal position size using Kelly criterion given 83% win probability and 1.7:1 reward-risk ratio.' The AI applies the Kelly formula and suggests: 'Optimal Kelly fraction: 0.145 (14.5% of capital). Recommended half-Kelly for safety: 7.25% = $3,625 position. At $510 per share: 7 shares.' This mathematical approach maximizes long-term growth while controlling risk of ruin.

Step 6: Monitor Positions and Execute Exits

Once in trades, track progress toward reversion targets. Ask 'Show me current P&L and distance to target for my cluster positions.' Sourcetable displays a dashboard with each position's entry price, current price, unrealized P&L, distance to cluster center (your target), and time in trade. You see at a glance which positions are moving as expected and which may need adjustment.

Set dynamic alerts for exit conditions: 'Notify me when NVDA reaches within $2 of cluster center or time in trade exceeds 10 days.' The AI monitors continuously and alerts you when conditions trigger. You can even ask for exit recommendations: 'Should I close my NVDA position now?' The system evaluates current deviation (now at 0.8 SD), time in trade (7 days), and remaining edge, responding: 'Position has captured 68% of expected reversion. Remaining edge is small. Consider closing 50-75% to lock profits while leaving runner for full target.'

Step 7: Analyze Performance and Refine Strategy

After accumulating trade history, ask Sourcetable to analyze your results: 'Show me performance metrics for all cluster trades this quarter.' Get comprehensive statistics: win rate, average win/loss, profit factor, maximum drawdown, Sharpe ratio, and comparison to buy-and-hold. The AI identifies patterns: 'Win rate is 78% overall but only 62% in trending markets. Consider filtering trades to range-bound conditions only.'

Request specific diagnostic analysis: 'Why did my TSLA cluster trade fail?' Sourcetable examines the trade context: 'Entry occurred during earnings week when implied volatility was 85th percentile. Historical data shows reversion probability drops to 54% during high-IV periods. Additionally, the cluster had only formed over 3 weeks—less established than your typical 6-week minimum. Suggest adding IV filter and cluster age requirements to entry rules.' This feedback loop continuously improves your strategy.

Real-World Applications of Cluster Analysis

Mean-reversion cluster strategies apply across multiple markets and trading styles. From day trading individual stocks to managing multi-asset portfolios, cluster analysis provides systematic entry and exit signals backed by statistical evidence. Here are specific scenarios where Sourcetable's cluster analysis delivers measurable trading edge.

Equity Day Trading and Swing Trading

Active traders use intraday and daily clusters to identify high-probability reversal points. You're watching AAPL which has formed a tight cluster between $178-$182 over the past month. News pushes the stock to $172 in morning trading—that's 2.4 standard deviations below the cluster center at $180. You ask Sourcetable 'What's the reversion probability for AAPL at $172?' and learn that in 15 historical instances of similar deviations, price returned to the cluster within 2 days 87% of the time with average gain of $6.50.

You enter long at $172 with 200 shares, risking $800 (stop at $168) to make $1,300 (target at $178.50, the lower cluster boundary). Sourcetable monitors the position and alerts you when price reaches $178.20 the next afternoon. You close for $1,240 profit—a 1.55:1 reward-risk ratio captured in 26 hours. The AI logs this trade and updates your strategy statistics, showing your cluster approach now has 76% win rate over 31 trades with 1.8 profit factor.

Currency Pairs and Forex Trading

Forex markets exhibit strong mean-reversion characteristics, making cluster analysis particularly effective. EUR/USD has traded in a cluster around 1.0850-1.0950 for six weeks. Central bank commentary drives a spike to 1.1050, putting price 2.1 SD above the cluster center. You ask Sourcetable 'Analyze EUR/USD cluster deviation and show reversion scenarios.' The AI reveals that at this deviation level, reversion to cluster center occurs within 5 days 81% of the time, with average retracement of 125 pips.

You short EUR/USD at 1.1045 with position size calculated by Sourcetable based on your 1.5% risk tolerance and 80-pip stop. The AI suggests 'With account size $100,000 and 1.5% risk ($1,500), position size is 1.875 mini lots (18,750 units) given 80-pip stop.' You execute the trade and set alerts for target at 1.0900 (cluster center) or stop at 1.1125. Three days later, price reverts to 1.0915 and you close for 130-pip gain, generating $2,437 profit. Sourcetable automatically updates your forex cluster statistics and identifies the next opportunity: GBP/JPY now showing 1.9 SD deviation.

Options Premium Selling at Cluster Boundaries

Combine cluster analysis with options strategies to enhance premium collection. NVDA has a well-defined cluster at $480-$500, and current price is $520—significantly extended. Instead of shorting stock, you sell out-of-the-money call spreads betting on reversion. Ask Sourcetable 'What strike prices optimize premium collection for NVDA reversion to $490 cluster?' The AI analyzes option chains and suggests: 'Sell $530/$540 call spread expiring in 21 days. Collect $3.20 premium. Probability of profit 76% based on cluster reversion patterns. Max risk $6.80 if price continues higher.'

You execute 10 spreads, collecting $3,200 premium with $6,800 max risk. Sourcetable monitors both the underlying cluster position and your option Greeks. Five days later, NVDA reverts to $495 and your spreads are worth $0.40. You close early for $2,800 profit (87% of max gain), avoiding the risk of holding to expiration. The AI notes: 'Early exit captured 87% of premium in 24% of time to expiration—efficient capital use. Annualized return on risk: 238%.' This quantified feedback helps you refine your options-cluster hybrid approach.

Portfolio Hedging with Cluster Divergence

Use cluster analysis to identify when your portfolio holdings have extended too far from equilibrium, signaling hedge opportunities. You hold long positions in tech stocks that have rallied strongly. Ask Sourcetable 'Show me cluster deviations for my portfolio holdings.' The AI reveals: 'MSFT: +2.3 SD, GOOGL: +1.9 SD, AMZN: +2.6 SD. Average portfolio deviation: +2.1 SD. Historical analysis shows 68% probability of 5%+ correction within 15 days when portfolio reaches this deviation level.'

Rather than selling winners and triggering taxes, you hedge with index put spreads or short-term tactical shorts in the most extended names. Sourcetable calculates hedge ratios: 'To hedge 50% of portfolio beta, short 120 shares of QQQ or buy 3 contracts of QQQ $380/$370 put spreads.' You implement the hedge and set alerts: 'Notify when average portfolio deviation returns below 1.0 SD.' Ten days later, tech pulls back 6%, your long positions decline $28,000, but your hedge gains $14,200. Net drawdown of $13,800 (49% hedged) versus $28,000 unhedged—cluster analysis helped you protect capital during the reversion.

Frequently Asked Questions

If your question is not covered here, you can contact our team.

Contact Us
What is the Ornstein-Uhlenbeck (OU) process and how is it used in equity mean reversion?
The Ornstein-Uhlenbeck process models a mean-reverting continuous-time stochastic process: dX = θ(μ - X)dt + σdW, where θ is the mean-reversion speed, μ is the long-run mean, σ is volatility, and W is Brownian motion. Applied to pairs or spread trading: fit the OU parameters to historical spread data to estimate (1) how quickly the spread reverts to mean (θ), (2) expected long-run value (μ), (3) typical deviation magnitude (σ). Trading signal: enter when spread deviates > 2 standard deviations from OU mean; exit at mean. OU model allows optimal entry/exit via half-life calculation: t½ = ln(2)/θ. A 5-day half-life means you expect 50% of the deviation to close in 5 days.
How do you find mean-reverting stock clusters for pair and basket trading?
Cluster identification methods: (1) Sector/industry grouping—stocks in the same GICS sub-industry often share common cost and revenue drivers, making their ratio stationary. (2) Cointegration testing—Engle-Granger or Johansen test. Two stocks are cointegrated if their linear combination is stationary (ADF test p-value < 0.05). (3) Correlation filtering—correlation > 0.7 over 252 days is necessary but not sufficient for cointegration. (4) Principal component analysis (PCA)—identify stock groupings with common factor exposures; pairs from same PCA cluster are mean-reverting candidates. (5) K-means clustering on fundamental characteristics (sector, size, leverage)—stocks in same cluster have higher cointegration probability.
What half-life of mean reversion is optimal for practical trading?
Optimal half-life range: 5-20 trading days. Below 5 days: extremely high turnover (200%+ monthly), transaction costs typically exceed profits. Above 20 days: slow convergence means you hold positions through significant market events; drawdowns from false signals become severe. The sweet spot (10-15 day half-life) allows: (1) Entry at 2σ deviation, (2) Expected reversion to mean in 10-15 days, (3) Exit at mean for 1-2% profit per trade, (4) Approximately 10-15 round-trip trades per pair per year. For a 20-pair portfolio with 1.5% average profit and $1M capital: annual gross return ~15-20%. After transaction costs (~0.3% round-trip for liquid ETFs): net 12-17%.
How do you test a pairs trade for statistical significance before trading it?
Statistical validation protocol: (1) Cointegration test—Engle-Granger: regress Stock A on Stock B, test residuals for stationarity with ADF test. p-value threshold: < 0.05 indicates cointegration. (2) Half-life estimation—fit OU process to residuals, calculate θ, compute t½ = ln(2)/θ. Accept if 5-20 days. (3) Hurst exponent—H < 0.5 indicates mean-reverting regime; H = 0.5 is random walk; H > 0.5 is trending. (4) Autocorrelation—lag-1 autocorrelation of spread should be negative (tendency to reverse). (5) Out-of-sample validation—test signals on last 12 months not used in parameter estimation. Common failure mode: passing in-sample tests but failing out-of-sample, indicating data snooping.
What position sizing approach is best for mean reversion cluster strategies?
Dollar-neutral pair position sizing: ensure each leg of the pair has equal dollar exposure. If Stock A is $50/share and Stock B is $100/share with hedge ratio 0.5 (from cointegration regression), buy 100 shares of A ($5,000) and short 25 shares of B ($2,500). Wait—dollar neutral requires buying $2,500 of A and shorting $2,500 of B. Adjust for beta: if A has beta 1.2 and B has beta 0.8, position size adjustment: A_shares/B_shares = (beta_B/beta_A) × (price_B/price_A) to achieve market-neutral as well as dollar-neutral. Most practitioners also target a fixed dollar risk per position: set 2% portfolio stop-loss per pair, size accordingly.
How does mean reversion clustering differ in different market regimes?
Mean reversion strategies work best when: (1) Market volatility is elevated (VIX 20-35)—increased noise creates temporary mispricings. (2) Sector-specific events cause transient dislocations. (3) Correlations within clusters are stable and high (low macro uncertainty). Mean reversion fails when: (1) One stock in a pair has a fundamental catalyst (earnings beat, M&A) that permanently changes the spread. (2) High macro uncertainty (VIX > 40)—correlations break down and cointegration relationships fail. (3) Sector structural change (airlines during COVID, retail vs e-commerce)—spreads may not revert because the relationship has permanently changed. Risk management: monitor z-score limits; if spread reaches 4σ from mean, cut loss rather than averaging in.
What annual return and Sharpe ratio can a properly implemented mean reversion strategy achieve?
Academic and practitioner benchmarks: (1) Simple pairs trading—2-4% annual return with Sharpe 0.5-0.8 (Gatev, Goetzmann & Rouwenhorst, 2006). Returns have declined from 11% in the 1990s as the strategy became crowded. (2) Cointegration-based pairs—5-8% annual return with Sharpe 0.7-1.0 using proper statistical methods. (3) Basket/cluster trading with 20+ pairs—better diversification reduces vol; Sharpe 0.8-1.2, annual returns 8-12%. (4) ETF pairs (more liquid, lower transaction costs)—3-6% annual with Sharpe 0.6-0.9. Key driver of declining returns since 2000: more quantitative capital chasing the same opportunities. Proprietary edge comes from (1) better cluster identification, (2) faster execution, (3) adding fundamental filters to improve signal quality.
Andrew Grosser

Andrew Grosser

Founder, CTO @ Sourcetable

Sourcetable is the AI-powered spreadsheet that helps traders, analysts, and finance teams hypothesize, evaluate, validate, and iterate on trading strategies without writing code.

Share this article

Sourcetable Logo
Ready to implement the Mean Reversion Clusters strategy?

Backtest, validate, and execute the Mean Reversion Clusters strategy with AI. No coding required.

Drop CSV