Half-Life Discounting of Old Regime Data

Suppose predictive relevance decays with a half-life of 6 months. Relative to current-regime data, what weight should you attach to observations that are 18 months old under this exponential-decay heuristic?