Asset Korelācija Kalkulators

A correlation of 0.7 means the two assets tend to move in the same direction about 70% of the time (informally) and that approximately 49% (0.7²) of the variance of one asset can be explained by linear movements in the other. In portfolio terms, a 0.7 correlation between two assets provides much less diversification benefit than a 0.0 or negative correlation. Adding a second asset with 0.7 correlation to an existing position reduces portfolio standard deviation only modestly compared to holding either asset alone. True diversification requires correlations closer to 0 or negative.

Correlations are not structural constants — they reflect the current economic and financial environment. In bull markets with stable macro conditions, risk assets can have relatively low correlations because company-specific factors dominate. In bear markets and crises, macro risk factors dominate (flight to quality, forced deleveraging, credit tightening), driving correlations toward 1.0 among risk assets. Additionally, increasing globalization, synchronized central bank policy, and quantitative easing have structurally increased cross-asset correlations over the past two decades. Risk models that use long-run average correlations systematically understate crisis risk because they don't capture this correlation instability.

With limited historical data (T observations) and many assets (N), sample correlation matrices become poorly estimated and noisy. When T < N, the sample matrix is not even invertible. Even when T > N, many estimated pairwise correlations will be far from their true values due to estimation error. Shrinkage estimators (Ledoit-Wolf, Oracle Approximating Shrinkage) regularize the sample correlation matrix by shrinking it toward a structured target (typically the identity matrix or a factor model). This reduces estimation error at the cost of introducing some bias, but produces better-behaved, positive-definite matrices that improve portfolio optimization and risk model stability.

Tail dependence measures the probability that two assets simultaneously experience extreme moves, beyond what linear correlation would predict. Two assets can have zero linear correlation but high tail dependence (they crash together even though they don't normally co-move). Copula models separate the marginal distribution of each asset from the dependence structure. The t-copula, unlike the Gaussian copula, exhibits symmetric upper and lower tail dependence — assets modeled by a t-copula are more likely to crash or spike together than a Gaussian copula predicts. This distinction was critical in the 2008 CDO crisis, where Gaussian copula models dramatically underestimated joint default probabilities under stress.

A rough rule of thumb: you need at least T > 3N observations (time periods vs. number of assets) for reasonably stable correlation estimates. With 252 daily observations (1 year), you can reliably estimate correlations for roughly 80 assets. For 2,000 assets (a typical large portfolio), you would need over 6,000 daily observations (24 years) — clearly impractical. This is why factor models (CAPM, Fama-French) are used for large portfolios: they decompose returns into a small number of common factors plus idiosyncratic risk, dramatically reducing the number of parameters to estimate while providing a more structured and stable covariance model.

The Diversification Ratio = (Weighted average individual volatility) / (Portfolio volatility). A ratio of 1.0 means no diversification benefit (all assets perfectly correlated). A ratio of 2.0 means portfolio volatility is half the weighted-average individual volatility — substantial diversification. Risk parity portfolios maximize the Diversification Ratio by weighting assets inversely proportional to their contribution to portfolio risk, rather than inversely proportional to expected return (as mean-variance optimization does). The Most Diversified Portfolio (MDP) — developed by Choueifaty and Coignard — selects weights that maximize the Diversification Ratio, making no assumptions about expected returns.

Normal VaR and portfolio optimization should use correlations estimated from a full market cycle or with exponentially weighted moving average (EWMA) smoothing that emphasizes recent data. For stress testing and crisis CVaR, use correlations estimated from historical crisis periods — specifically from windows that include the GFC (2008), COVID crash (2020), or other relevant stress events. Crisis-period correlations among risk assets are typically 30–50% higher than full-period averages. Using stressed correlations in capital adequacy analysis prevents the dangerous assumption that diversification benefits observed in calm markets will persist in crisis scenarios.