Durbin-Watson Autocorrelation Test Fundamentals
📌 The Durbin-Watson (DW) test is a classic assumption test used in regression analysis specifically for time series data.
📊 A good regression model must pass the autocorrelation test criteria.
📉 The test detects the presence or absence of autocorrelation symptoms.

EViews Practical Application
💾 The example data utilized includes two independent variables ($\text{X}_1, \text{X}_2$) and one dependent variable ($\text{Y}$) spanning from 2013 to 2020 (quarterly data).
💻 To input the data in EViews, set the structure type to Quarterly, starting in 2013 and ending in 2020.
⚙️ After estimating the model, the Durbin-Watson statistic ($\text{DW}$) obtained from the output for the sample data was 1.745973.

Durbin-Watson Interpretation Criteria
⚖️ Autocorrelation Present (Fail Test): If $\text{DW} < \text{DL}$ or $\text{DW} > 4 - \text{DL}$, the data exhibits autocorrelation.
✅ No Autocorrelation (Pass Test): If $\text{DU} < \text{DW} < 4 - \text{DU}$, the data passes the test (no autocorrelation).
❓ Inconclusive: If $\text{DW}$ falls between $\text{DL}$ and $\text{DU}$, or between $4 - \text{DU}$ and $4 - \text{DL}$, no definitive decision can be made.

Analysis of Sample Output
🔢 The sample involved $n = 32$ observations and $k = 2$ independent variables, tested at a 5% significance level ($\alpha = 0.05$).
📉 Based on the Durbin-Watson reference table for $n=32$ and $k=2$: $\text{DL} = 1.3093$ and $\text{DU} = 1.5736$.
🔄 The critical boundaries calculated were $4 - \text{DL} = 2.6907$ and $4 - \text{DU} = 2.4264$.
🌟 Conclusion for Sample Data: Since the calculated $\text{DW}$ statistic ($\text{1.745}$) falls between $\text{DU}$ ($\text{1.5736}$) and $4 - \text{DU}$ ($\text{2.4264}$), the conclusion is that no autocorrelation is present, and the data passes the test.

Key Points & Insights
➡️ If autocorrelation is detected (data fails the test), transformation methods such as $\text{Cochrane-Orcutt (CO)}$ or $\text{Prais-Winsten (PW)}$ should be considered.
➡️ If the result is inconclusive, the recommendation is to use alternative methods like the Lagrange Multiplier (LM) test (or $\text{Breusch-Godfrey test}$).
➡️ The values for $\text{DL}$ and $\text{DU}$ must be sourced from the official Durbin-Watson reference table corresponding to the sample size ($n$) and number of predictors ($k$).

📸 Video summarized with SummaryTube.com on Nov 09, 2025, 14:40 UTC