Intermarket Correlation Trading

Directional trading has higher reward potential but also higher risk. Spreads offer: 1) reduced directional risk - a market crash affects both legs. 2) More consistent returns - spread behaviour is more predictable than absolute returns. 3) A statistical edge - spread mean reversion is a documented phenomenon. 4) Lower drawdowns - spread volatility is typically lower. Trade-off: you sacrifice maximum upside for more consistent, lower-risk returns.

Several methods: 1) trading and charting platforms (e.g. TradingView) may have correlation tools. 2) Spreadsheet: download daily closes and calculate a 30-day rolling correlation using the CORREL function. 3) Python/R scripts for real-time calculation. 4) Overlay charts for a visual sense of correlation. Start with daily monitoring; advance to intraday as you develop. Track correlation in a journal alongside your spread trades.

This happens when correlation breaks down - both instruments move against you simultaneously. Example: long financials (which falls), short resources (which rises) - lose on both. Protection: 1) always use stop losses on spread positions. 2) Reduce size during high A-VIX (breakdown risk). 3) Monitor correlation daily - exit if correlation drops significantly. 4) Accept this as a risk of the strategy - proper sizing ensures no single breakdown is catastrophic.

More than single-instrument trading due to multiple positions. Minimum practical: ~A$40,000-60,000. This allows positions in a couple of instruments (e.g. an SPI 200 or Mini SPI 200 leg plus a sector-ETF or LEPO leg) plus a buffer for drawdowns. Better: A$80,000-100,000 for proper position sizing and diversification across 2-3 spreads. The capital requirement is a barrier but also a safety feature - undercapitalised correlation trading is very risky.

Yes, correlations can shift due to structural changes. Examples: regulatory changes affecting one sector, a new index composition, economic shifts (a sustained commodity-cycle change can reset the banks-miners relationship). Signs of permanent change: 1) the correlation stays at a new level for 3+ months. 2) A fundamental reason exists. 3) The spread doesn't revert despite extreme readings. Adaptation: periodically recalculate normal correlation levels. If the relationship has structurally changed, update your spread models accordingly.

Use correlation for: shorter-term trades (intraday to a few days), simpler implementation, when instruments have a stable correlation. Use cointegration for: longer-term trades (weeks), when you need statistical rigour, when correlation alone seems insufficient. Practical approach: start with correlation-based spreads for simplicity. Graduate to cointegration analysis as you develop skills. For most retail traders, correlation-based spreads are sufficient and easier to manage.

It depends on the holding period and market conditions. Guidelines: intraday spreads - calculate at the start of the day, no intraday adjustment. Multi-day spreads - recalculate weekly or when beta changes >5%. High-volatility periods - recalculate more frequently (every 2-3 days). Stable periods - weekly is sufficient. Balance: too-frequent adjustment = high costs; too-infrequent = tracking error. Monitor the spread residual - if it's drifting, the ratio may need adjustment.

Options from best to acceptable: 1) API-based execution: program both legs to send simultaneously - fastest, most reliable. 2) Multi-leg orders: some platforms allow them. 3) Two terminals: have both legs ready, execute within seconds of each other - manual but workable. 4) Execute the less liquid leg first, then immediately execute the liquid leg. Slippage budget: expect 0.02-0.05% slippage on each leg. Factor this into trade-expectancy calculations.

Key differences: 1) single-stock correlations are generally lower and less stable than index correlations. 2) Stock-specific events (results, news) can break correlations temporarily. 3) Liquidity varies more for stocks - execution risk is higher. 4) Fundamental divergences are more common. Approach: use same-sector pairs (e.g. CBA-NAB within banks, BHP-RIO within miners) rather than cross-sector (banks-miners), which is genuinely divergent in Australia. Monitor upcoming events for both legs. Require a higher correlation threshold (0.8+ for same-sector pairs vs 0.75 for indices). Use wider stops to accommodate higher variability.

The A-VIX is a crucial correlation-regime indicator: low A-VIX (<11): stable correlations, spread strategies work well, normal sizing appropriate. Medium A-VIX (11-16): monitor more closely, correlations may weaken, consider slightly reduced size. High A-VIX (>16): correlations unstable, can spike or break down, reduce spread exposure significantly. An A-VIX spike (>20): high correlation-breakdown risk, consider pausing spread strategies. Rule of thumb: for every 5 points of A-VIX above ~13, reduce spread position size by 20%. (The A-VIX generally runs lower than other volatility indices, so the thresholds sit lower.)

Kalman filter implementation: 1) State equation: the hedge ratio follows a random walk (beta_t = beta_{t-1} + noise). 2) Observation equation: spread = y - beta x x + error. 3) Update: as each new price pair arrives, update the beta estimate and its uncertainty. 4) Libraries: pykalman in Python, or implement from scratch using the standard Kalman equations. Benefits: smoother adaptation than rolling regression, weights recent data appropriately. Challenges: you need to tune the noise parameters, and it can be unstable if poorly calibrated. Validate against simple rolling regression before production use.

Primary risks: 1) systemic correlation breakdown: during crises, all spreads may blow out simultaneously despite individual diversification. 2) Hidden correlations: spreads you thought were independent may be correlated through hidden factors (in Australia, many large caps share China/commodity exposure). 3) Liquidity clustering: multiple legs may become illiquid simultaneously. 4) Model risk: if cointegration relationships change, the entire portfolio strategy fails. Mitigation: stress test the portfolio under historical crises, monitor inter-spread correlations, maintain a cash buffer for margin calls, and have a hard portfolio-level stop loss.

Components: 1) Data pipeline: real-time tick data -> aggregated bars -> database storage. 2) Calculation engine: parallel processing of multiple correlation/spread calculations. 3) Alert system: threshold triggers -> notification (SMS, email, push). 4) Dashboard: real-time visualisation of correlations, spreads, Z-scores. 5) Logging: all calculations and alerts logged for analysis. 6) Backtesting integration: the ability to replay historical data. Tech stack: Python/pandas for calculations, Redis/Kafka for real-time data, PostgreSQL for storage, Grafana for visualisation. Start simple, add complexity as needed.

Testing hierarchy: 1) Augmented Dickey-Fuller (ADF): the standard stationarity test, p < 0.05 preferred. 2) KPSS test: complementary to ADF (null hypothesis is stationarity), confirms the ADF result. 3) Phillips-Perron: robust to serial correlation and heteroskedasticity. 4) Hurst exponent: H < 0.5 indicates mean reversion, H > 0.5 indicates trending. Best practice: require ADF p < 0.05 AND KPSS p > 0.05 AND Hurst < 0.45 for high-confidence stationarity. Re-test monthly, as relationships can change (cross-sector spreads especially).

Portfolio allocation framework: 1) correlation/spread strategies: 20-40% of trading capital (their lower volatility allows a larger allocation). 2) Within the correlation allocation: no single spread > 30% of correlation capital. 3) Correlation with the rest of the portfolio: ensure correlation strategies don't simply replicate directional bets elsewhere. 4) Margining: correlation trades often get a spread-margin benefit - factor this into capital efficiency. 5) Drawdown budget: allocate a specific drawdown budget to correlation strategies (e.g., max 5% portfolio drawdown from spread trades). 6) Rebalancing: monthly rebalance between correlation and other strategies based on performance and regime.