Skip to content

Further Extensions & Use Cases

Example: Bivariate \(MA(1)\) process

A bivariate Moving Average process of order 1 involves two interrelated series. Consider the following specification:

\[ \begin{pmatrix} Y_{1t} \\ Y_{2t} \end{pmatrix} = \begin{pmatrix} e_{1t} \\ e_{2t} \end{pmatrix} - \begin{pmatrix} 0.2 & -0.4 \\ -0.2 & 0.6 \end{pmatrix}\begin{pmatrix} e_{1,t-1} \\ e_{2,t-1} \end{pmatrix};\quad \Sigma = \begin{pmatrix} 4 & 1 \\ 1 & 4 \end{pmatrix} \]

To analyze the dependence structure, we compute the following matrices:

  • \(\Gamma(0)\): The variance-covariance matrix of the process, calculated as \(\Gamma(0) = \Sigma + \Theta_{1} \Sigma \Theta_{1}'\).
  • \(\Gamma(1)\): The cross-covariance matrix at lag 1, calculated as \(\Gamma(1) = -\Sigma \Theta_{1}'\).
  • \(\Gamma(l)\) for \(l \geq 2\): For an \(MA(1)\) process, all covariances at lags greater than the order \(q\) are zero (\(\Gamma(2+) = 0\)).
  • \(\rho(0)\) and \(\rho(1)\): The contemporaneous and lag-1 correlation matrices, derived by normalizing the covariances.

Example: \(VAR(p)\) process

The Vector Autoregressive process of order \(p\) is defined as:
$\(\Phi(B)Y_{t} = \delta + e_{t}\)$

Where:
* \(\Phi_{i}\): \(k\)-dimensional square matrices of parameters.
* \(e_{t}\): \(k\)-dimensional vector of residuals (a purely random process/Vector White Noise).
* \(\delta\): A vector of constants.

Duality of Representations:
* \(VMA(1) \equiv VAR(\infty)\): An invertible Vector Moving Average process can be expressed as an infinite Autoregressive process.
* \(VAR(p) \equiv VMA(\infty)\): A stationary Vector Autoregressive process can be expressed as an infinite Moving Average process of random shocks.

Further Observations

  • We continue to look at Covariance Matrices and Correlation Matrices to understand the system.
  • The underlying expressions and formulae remain conceptually identical to univariate cases, but we utilize the corresponding multivariate notation and forms (vectors and matrices) to account for inter-series dependencies.

Application Areas

Macroeconomics

  1. Monetary Policy Analysis: Examining how interest rate changes, inflation, and monetary policies impact GDP, unemployment, and exchange rates.
  2. Economic Forecasting: Using VARMA to forecast GDP growth and employment while accounting for the complex interactions between these variables.

Supply Chain and Operations

  1. Inventory Management: Improving inventory levels to reduce stockouts and excess inventory by understanding interrelated demand.
  2. Logistics and Shipping: Forecasting shipping demand and lead times by taking interrelated demand factors into account.

Financial Markets

  1. Asset Pricing: Analyzing the co-movements between different assets to price risk.
  2. Volatility Forecasting: Modeling how volatility in one asset or market spills over into another.

Energy Markets

  1. Electricity Load Forecasting: Predicting demand to prevent system overloads.
  2. Oil and Gas: Modeling price and supply dependencies across different energy sectors.

Healthcare

  1. Epidemiology: Tracking the spread of diseases across different regions or demographics simultaneously.
  2. Hospital Resource Planning: Forecasting patient loads across different departments to optimize resource allocation.

Climate Science

  1. Temperature & Weather Forecasting: Jointly modeling variables like pressure, temperature, and wind.
  2. Air Quality Monitoring: Analyzing the relationship between different pollutants (e.g., \(PM_{2.5}\), \(NO_{2}\)) across various geographical sensors.