- Time series
- Time series: observations of one variable in time order, such as daily returns or a monthly spread.
- Stationary
- Stationary (weakly): the mean and variance do not change over time, and the covariance between two values depends only on how far apart they are.
- Lag
- Lag k: the value k periods earlier. X_{t−1} is the first lag of X_t.
- Autocorrelation function (ACF)
- Autocorrelation function (ACF): the correlation ρ_k between a series and itself k periods earlier, plotted against k.
- ±1.96/√n band
- The ±1.96/√n band: the range that about 95% of sample autocorrelations of white noise fall in, for a series of length n.
- AR(1)
- AR(1): X_t = c + φX_{t−1} + ε_t, with independent shocks ε_t. Stationary when −1 < φ < 1.
- Mean reversion
- Mean reversion: a tendency to move back towards a long-run mean; under AR(1) a gap shrinks by the factor φ each period.
- Half-life
- Half-life: the time for half of a gap from the mean to disappear, ln 0.5 / ln φ periods under AR(1).
- White noise
- White noise: a series of uncorrelated values with mean 0 and constant variance; all its autocorrelations are 0.
- Random walk
- Random walk: X_t = X_{t−1} + ε_t. Shocks never fade, its variance grows with time, and it is not stationary.
- Ljung-Box test
- Ljung-Box test: Q(h) = n(n + 2) Σ ρ̂_k²/(n − k) over lags 1 to h, compared with χ² with h degrees of freedom.