Description

Here is a 10-line summary of the provided PDF on State Space Models and Markov Switching:

1. **State Space Framework:** The chapter introduces State Space Models (SSM), which decompose a time series into a state equation (describing the evolution of unobserved components) and an observation equation (linking those states to observed data).
2. **Unified Representation:** SSM provides a unified framework for ARMA, ARIMA, and structural time series, with advantages including natural handling of missing data and optimal forecasting via the Kalman Filter.
3. **ARMA(1,1) in State Space:** It demonstrates algebraically that an ARMA(1,1) model can be re-written as a specific state space form, proving their equivalence through substitution steps involving the state variable \(X_t\).
4. **Kalman Filter Equations:** For time-invariant models, the chapter derives the full Kalman filter recursions, including formulas for innovation, variance (\(V_t\)), filtering (\(X_{t|t}\)), forecasting (\(X_{t+1|t}\)), and smoothing (\(X_{s|t}\)).
5. **SARIMAX Application:** Using real GDP, consumption, and investment data, the text fits a SARIMAX(1,0,0)(0,0,0)4 model, applying a log transformation and using PACF to justify the AR(1) component.
6. **Markov Switching Introduction:** Markov Switching Models (MSM) are introduced to handle regime changes, where parameters like the mean (\(\mu\)), AR coefficient (\(\beta\)), and variance (\(\sigma^2\)) depend on an unobserved state variable \(S_t\).
7. **DAX Returns with ARCH:** Fitting a 2-regime Markov switching model to DAX log returns reveals a low-volatility regime (\(\sigma^2 = 5.44\times 10^{-5}\)) and a high-volatility regime (\(\sigma^2 = 3.00\times 10^{-4}\)).
8. **Model Adequacy via Ljung-Box:** Residual analysis using the Ljung-Box test is crucial. For the Federal Funds Rate, simple 2-regime models fail as p-values are near zero, indicating significant leftover autocorrelation.
9. **Complex Model Specification:** To adequately model the Federal Funds Rate, a more complex 3-regime model with 4 regressors (including two lags of the dependent variable, output gap, and inflation) was required to achieve white noise residuals.
10. **Asset Returns Failure:** A 2-regime Markov model applied to absolute asset returns is deemed inadequate, as the smoothed probabilities show rapid, noisy switching with no persistent states, and residual diagnostics fail.