Quantitative Trading · Level II · Elective
Time-Series Econometrics
The statistical models behind systematic strategies: stationarity and unit roots, ARMA models, GARCH volatility, cointegration tests, Kalman filters and regime-switching models, each with what it is used for in trading and how it fails.
What you will be able to do
- Test for stationarity and choose models accordingly.
- Fit and interpret ARMA and GARCH models.
- Apply cointegration tests, Kalman filters and regime-switching models to trading problems.
Section 1 of 2
Stationarity and unit roots
A time series is (weakly) stationary if its mean, variance and autocorrelations do not change over time. Most statistical models assume it. Prices are usually not stationary: they wander like a random walk, which has a unit root. Returns usually are close to stationary in mean, though their volatility changes.
The augmented Dickey-Fuller (ADF) test checks for a unit root: its null hypothesis is that the series has one. A low p-value rejects the unit root, suggesting stationarity. The KPSS test reverses the null (stationarity), and using both gives a more balanced view.
Regressing one non-stationary series on another can produce a spurious relationship with a high R² even when the series are independent. This is why returns, not prices, are usually modelled, unless the prices are cointegrated.
Keep this in mind: Model stationary series; test for unit roots with ADF and KPSS; beware spurious regressions of levels.
Make the idea stick
Quick checks before you move on.
Quick check 1 of 2 · Stationarity and unit roots
What is the null hypothesis of the ADF test?
Quick check 2 of 2 · Stationarity and unit roots