This paper takes the Asian crisis as an example to show that the Autoregressive Conditional Hazard (ACH) model is a powerful tool for studying the time series features of speculative attacks. The ACH model proposes a duration variable to capture the changes in the frequency of attacks, which might be an important factor influencing investors' expectations. The empirical results show that the ACH model explains the crisis far better than the Probit model. The duration variable is highly significant while most fundamentals are not. The contagion effect is tested and accepted under the ACH specification.