The paper discusses the dangers of using correlation to measure dependence in financial assets. While correlation is a key parameter in modern portfolio theory, it is not always a reliable measure of dependence, especially when asset returns are not normally distributed. Real-world returns often exhibit negative skewness and excess kurtosis, making correlation an inadequate measure of dependence. Even when correlation is a valid measure, people often misunderstand its implications. For example, the correlation between hedge fund returns and stock market returns is higher in down markets than in up markets, which can be attributed to technicalities rather than actual dependence. Elliptical distributions, such as the normal distribution, are ideal for modeling dependence, but real-world distributions are rarely elliptical. This has led to the widespread assumption of normality in econometrics, which is increasingly being challenged. Conditional correlations, which involve splitting data based on variable characteristics, can show higher correlations during extreme events, but this is often due to technical factors rather than actual increased dependence. The paper highlights that conditional correlations can be misleading because they depend on the marginal distributions of variables. When the variance within a segment is high relative to the overall variance, conditional correlations can appear higher, even if the overall correlation remains constant. This suggests that extreme movements may not necessarily be more correlated than overall movements. The paper concludes that correlation is a limited tool for measuring dependence in real-world financial data, and more sophisticated methods are needed to understand true dependence structures. The findings emphasize the importance of moving away from the normality assumption and adopting more realistic models for financial risk management.The paper discusses the dangers of using correlation to measure dependence in financial assets. While correlation is a key parameter in modern portfolio theory, it is not always a reliable measure of dependence, especially when asset returns are not normally distributed. Real-world returns often exhibit negative skewness and excess kurtosis, making correlation an inadequate measure of dependence. Even when correlation is a valid measure, people often misunderstand its implications. For example, the correlation between hedge fund returns and stock market returns is higher in down markets than in up markets, which can be attributed to technicalities rather than actual dependence. Elliptical distributions, such as the normal distribution, are ideal for modeling dependence, but real-world distributions are rarely elliptical. This has led to the widespread assumption of normality in econometrics, which is increasingly being challenged. Conditional correlations, which involve splitting data based on variable characteristics, can show higher correlations during extreme events, but this is often due to technical factors rather than actual increased dependence. The paper highlights that conditional correlations can be misleading because they depend on the marginal distributions of variables. When the variance within a segment is high relative to the overall variance, conditional correlations can appear higher, even if the overall correlation remains constant. This suggests that extreme movements may not necessarily be more correlated than overall movements. The paper concludes that correlation is a limited tool for measuring dependence in real-world financial data, and more sophisticated methods are needed to understand true dependence structures. The findings emphasize the importance of moving away from the normality assumption and adopting more realistic models for financial risk management.