User:Dr. Ghulam Ghouse/sandbox

Ghouse Equation
In traditional Econometrics, non-stationarity is often seen as the main cause of spurious regression, and to address this, the common methods used are unit root and Cointegration analysis. However, other factors contributing to spurious regression haven't been thoroughly discussed. Ghouse's experiment shed light on this by suggesting that spurious regression occurs not only due to non-stationarity but also when relevant variables or lag values are missing in time series data. Ghouse's experiment introduces an alternative solution called the Ghouse Equation, which relies entirely on the autoregressive distributed lag mechanism. Unlike traditional approaches, this method doesn't require the assumption of pre-testing procedures like unit root testing or bound testing. This advantage makes the proposed methodology more efficient compared to existing cointegration procedures. Assuming that spurious regression happens due to non-stationarity and that unit root and cointegration testing are used to fix it, it is still challenging to draw reliable conclusions. The issue lies in the lack of a good size and power of unit root tests for small sample size. Unit root and cointegration procedures involve making several decisions beforehand, like choosing lag length,trend, and ensuring structural stability. Making decisions based on the data requires a series of tests, each with its specific statistical errors (type-I and type II errors). The combined probability of errors across all tests makes the results of the unit root test less dependable. Consequently, even after four decades, the literature remains underdeveloped and inconclusive due to these factors.

Granger and Newbold Experiment
Granger and Newbold in 1974 generated two time series variables $$x_t $$ and $$y_t $$, both of which are functions only of their own lags, and no other variable is involved in the construction of both variables.
 * $$y_t = y_t-1 + e_t $$
 * $$x_t = x_t-1 + e_t $$

They ran the following regression models:
 * $$y_t = a_0 + a_1x_t + e_t$$
 * $$x_t = b_0 + b_1y_t + e_t$$

The main aim of the experiment was to demonstrate that non-stationary time series causes spurious regression. However, this experiment holds several other implications that weren't explored. There was a notable discovery during the experiment: a surprisingly strong correlation between variables, despite their lack of actual relation. This resulted in spurious regression, yet the underlying cause was not thoroughly investigated. In the major part of the econometric literature, the focus primarily rested on non-stationarity as the root cause of this issue, while the problem related to missing variables (lags) was completely overlooked.

Ghouse Experiment
The Ghouse equation is methodically built upon the foundations of the Autoregressive Distributed Lag (ARDL) mechanism. The bound testing procedure also utilizes the ARDL mechanism. However, it extends its application by incorporating bound testing and, to some extent, unit root testing. One significant discovery from the Ghouse Experiment is that the Ghouse equation does not require any extensions like bound testing or the unit root procedure. This adaptation indeed positions it as the simplest version compared to the existing cointegration procedures.

Data Generating Process of Ghouse Equation

 * $$\begin{bmatrix} \mathbf{x_t} \\ \mathbf{y_t} \end{bmatrix} =

\begin{bmatrix} \mathbf{\theta_{11}} & \mathbf{\theta_{12}} \\ \mathbf{\theta_{21}} & \mathbf{\theta_{22}} \end{bmatrix} \begin{bmatrix} \mathbf{x_{t-1}} \\ \mathbf{y_{t-1}} \end{bmatrix} + \begin{bmatrix} \mathbf{a_1} & \mathbf \\ \mathbf & \mathbf \end{bmatrix} \begin{bmatrix} \mathbf{1} \\ \mathbf{t} \end{bmatrix} + \begin{bmatrix} \mathbf{e_{xt}} \\ \mathbf{e_{yt}} \end{bmatrix} $$

Where the error structure is:



\begin{bmatrix} \mathbf{e_{xt}} \\ \mathbf{e_{yt}} \end{bmatrix} \sim\ \mathcal{N} (\begin{bmatrix} \mathbf{0} \\ \mathbf{0} \end{bmatrix}\begin{bmatrix} \mathbf{1} & \mathbf{\rho} \\ \mathbf{\rho} & \mathbf{1} \end{bmatrix}) $$ It can be rewritten as for simplification of notation:
 * $$Y_t = AY_{t-1}+ Bd + e_t$$