Skip to main content

Correlation & Concentration Risk

Updated Jun 23, 2026 at 8:47pm

Research Draft Medium 1,157 words

Per-trade risk control (e.g. risking 1% of equity per position) is necessary but not sufficient. Correlation & concentration risk is the portfolio-level failure that occurs when several positions, each sized to a small independent risk, are actually driven by the same underlying factor — the same sector, theme, macro variable, or style. When that hidden link is present, the trades are not five separate 1% bets; they are one 5% bet wearing five tickers. The core tension is that a book can look diversified on a position-count basis while being highly concentrated on a risk basis — and the moment that matters most (a sharp sell-off) is precisely the moment correlations rise toward 1 and the hidden concentration is revealed as a single coordinated loss.

How it's calculated / measured

Two distinct quantities matter, and they are easy to conflate:

  • Pairwise correlation (ρ) between two positions' returns, ranging −1 to +1. It measures how much two names tend to move together. Correlation is estimated over a lookback window (e.g. 60–120 trading days), so it is inherently backward-looking and regime-dependent — the number you compute in a calm market is not the number you will experience in a crash.
  • Portfolio heat — the sum of open risk across all live positions: the total percentage of account equity you would lose if every open trade hit its stop simultaneously. If you hold five trades each risking 2%, naive portfolio heat is 10% (Pro Trader Dashboard; Van Tharp Institute summaries).

The link between them: correlation converts portfolio heat from a worst-case-if-everything-goes-wrong figure into a base-case figure. For independent positions, all stops triggering at once is improbable. For positions correlated near +1, all stops triggering together is the expected behavior on a bad day — so the effective heat of a correlated cluster approaches the simple sum of its parts. The practical heuristic that follows is to treat a correlated cluster as a single risk unit: count the whole oil-stock basket as one position's worth of heat, not three.

How it's used in practice

The discipline has three standard levers, all style-agnostic:

1. Cap total portfolio heat. Set a maximum on the sum of open risk across the book. Commonly cited rules of thumb put this in the 6–10% range for retail traders (Van Tharp's "never more than 6–10% at risk at once"; Pro Trader Dashboard cites 5–10% retail, 3–6% institutional). The exact cap is a risk-tolerance choice, not a law — the principle (cap the sum, not just the parts) is what's universal. 2. Limit positions per sector / theme / factor. Constrain how many correlated names can be open at once (e.g. no more than N positions in one sector, or a hard cap on combined exposure to a single theme). This directly attacks hidden concentration. 3. Size down when the whole book is one macro bet. If a scan reveals that the open positions are all long the same factor (high-beta growth, rate-sensitives, a single commodity), the book is that macro bet — reduce per-position size or trim names so the cluster's combined risk fits the heat budget for a single idea.

A trader running five long tech names in a momentum tape does not hold a diversified book; they hold a leveraged bet on tech momentum. The job of this risk layer is to make that fact visible before the drawdown, and to force the size down to match.

Standing & evidence

The crisis-correlation phenomenon is one of the better-documented findings in portfolio risk, not folklore:

  • Left-tail correlations are systematically higher than right-tail correlations. Page & Panariello ("When Diversification Fails," Financial Analysts Journal, 2018) show that across styles, sizes, geographies, and alternative assets, correlations conditioned on downside events are much higher than those conditioned on upside events — diversification weakens exactly when you need it. This is the rigorous academic anchor for the claim.
  • Measured crisis spikes. Industry analyses report average pairwise equity correlations rising from a calm-market range of roughly 0.40–0.50 to the 0.8–0.9+ range during stress — figures commonly cited for the 2008 crisis (~0.80+), the euro crisis, and the March 2020 COVID sell-off (Cambridge Associates; Morningstar's "Correlations Going to 1"). Treat the exact decimals as illustrative magnitudes, not precise constants — they vary by universe and estimation method.
  • Mechanism. In a panic, selling becomes indiscriminate and liquidity-driven rather than fundamentals-driven; correlated forced selling (deleveraging, margin calls, redemptions) pushes otherwise-unrelated assets down together. The result: the diversification you measured in calm markets does not exist in the market you actually need it for.

Strengths & limitations

Strength: Portfolio heat plus a correlation lens is a cheap, robust budgeting discipline. It does not require predicting anything — it just stops you from unknowingly running 5× the risk you think you're running, and it imposes a hard ceiling on the worst case.

Limitations — be honest about them:

  • Correlation is unstable and backward-looking. A low historical correlation can betray you when a new common driver appears (a macro shock, a sector rotation). The estimate is most reliable in the regime you measured it in, and least reliable in the regime that hurts you.
  • It is risk-budgeting, not a precise model. Treating a cluster "as one unit" is a deliberately conservative heuristic, not a calibrated number. Don't dress it up as VaR-grade precision.
  • The single most common misuse is counting positions instead of risk — assuming "I hold eight different stocks, so I'm diversified" when all eight share one factor. Diversification is a property of the risk drivers, not the ticker count.
  • Correlations near +1 in tail events mean any cross-position offset you were relying on (one name hedging another) can vanish precisely in the sell-off. Size for the world where ρ → 1, not the world where ρ = 0.

System relevance

This node sits at the portfolio level and is the layer above per-trade controls. It cross-links directly to:

  • Position sizing & R-multiples (sibling risk nodes) — those define risk per trade (the 1R unit); this node aggregates those units and applies the correlation/heat ceiling on top. A correct per-trade size can still produce an over-risked book.
  • The Delvantic analysis pipeline / Augustus trade-setup layer should treat correlated open positions as a single risk unit when checking a new candidate against the heat budget, and reduce or reject size when the book is already concentrated in the candidate's sector/theme/factor. The hard caveat to pass downstream: historical correlation underestimates crisis correlation — budget for ρ → 1 in stress, not the calm-market estimate.

Sources