The Impact of Model Instability on Long-Term Investors

Should long-term investors account for time-variation in model parameters? We develop a time-varying Vector Autoregressive model that can handle time-variation in intercepts, slopes, volatility and correlation, the leverage effect in volatility and fat tails. Long-term investors should take time-variation in volatility and correlation into account due to its high persistence, but can safely ignore time-variation in slope coefficients and intercepts due to its low persistence. Modeling time-variation in parameters with a random walk specification or a specification that pools persistence across parameters leads to a large overestimation of the riskiness of stock returns and an underinvestment in the stock market