Scribd
Upload
8 views3 pages

Asg 3

The document discusses various statistical models and tests related to autocorrelation in regression analysis, emphasizing the weak form of market efficiency. It presents examples including logarithmic regression, adjustments for AR(1) autocorrelation, partial adjustment models, and common factor tests, highlighting the implications of autocorrelation on OLS estimates. Key findings indicate the presence of autocorrelation in certain models and the effectiveness of transformations to address this issue.

Uploaded by

Jay Vora
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
8 views3 pages

Asg 3

The document discusses various statistical models and tests related to autocorrelation in regression analysis, emphasizing the weak form of market efficiency. It presents examples including logarithmic regression, adjustments for AR(1) autocorrelation, partial adjustment models, and common factor tests, highlighting the implications of autocorrelation on OLS estimates. Key findings indicate the presence of autocorrelation in certain models and the effectiveness of transformations to address this issue.

Uploaded by

Jay Vora
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Part 1

Correlation between rt and rt-1 is 0.038.

Under the weak form of market efficiency, past price information should not be useful for
predicting future prices. A correlation close to zero supports this hypothesis, suggesting that
historical returns do not provide a systematic basis for forecastable profit opportunities.
Excel sheet attached.

Part 2
Example 1: Logarithmic Regression of Housing Services
This example involves a regression where the dependent variable is the logarithm of housing
services (LGHOUS), and the independent variables are:
• LGDPI: The logarithm of disposable personal income.
• LGPRHOUS: The logarithm of the relative price of housing services.
Key Results:

• The regression output shows a Durbin-Watson statistic of 0.85, which is well below 2.
This indicates positive autocorrelation in the residuals.

• For a 1% significance level, with 2 explanatory variables and 36 observations, the


critical values for the Durbin-Watson test are:

• dL=1.15

• dU=1.65
• Since d=0.85<dL , we reject the null hypothesis (H0: ρ=0) and conclude that positive
autocorrelation is present.
Implications:

• Positive autocorrelation means that residuals from one period are correlated with
residuals from previous periods.
• This violates the Gauss-Markov assumption that error terms should be independent,
leading to inefficient OLS estimates and potentially biased standard errors.

Example 2: Adjusting for AR (1) Autocorrelation


To address autocorrelation, the model was transformed to include an AR (1) process for the
error term: ut=ρut−1+εt where εt represents a random error term (white noise).
Transformation Process:

1. The regression equation was lagged by one period and multiplied by ρ.


2. Subtracting this lagged equation from the original equation resulted in a transformed
model: Yt−ρYt−1=β1(1−ρ) +β2Xt−β2ρXt−1+εt
3. This transformation eliminates autocorrelation by reducing the disturbance term to
white noise (εt).
Results:

• The nonlinear specification was estimated using maximum likelihood methods.

• The estimate of ρ was very high (0.97), indicating severe autocorrelation in the
original model.

• After adjusting for autocorrelation, the Durbin-Watson statistic improved


significantly, confirming that the transformed model was free from autocorrelation.
Example 3: Partial Adjustment Model

This example introduces a partial adjustment model, which includes a lagged dependent
variable (Yt−1) as an explanatory variable: Yt=β1+β2Xt+β3Yt−1+ut
Key Issues:

• If ut is subject to AR (1) autocorrelation (ut=ρut−1+εt), OLS estimates become


inconsistent because Yt−1 (a regressor) is correlated with ut, violating Gauss-Markov
assumptions.
Detection and Results:

• The Durbin-Watson test is invalid in this case because it tends to be biased toward 2
when lagged dependent variables are included.
• Instead, the Durbin h-statistic was used to detect autocorrelation. In this case:
• h=0.86, which is below the critical value of 1.96 at a 5% significance level.

• Thus, we fail to reject the null hypothesis of no autocorrelation in this


adjusted model.

Example 4: Common Factor Test


This example evaluates whether an AR (1) model is an appropriate specification by
comparing it to a more general ADL (1,1) model (autoregressive distributed lag model).

Models Compared:
1. AR (1) Model: Yt=β1(1−ρ) +ρY(t−1)+β2Xt−β2ρX(t−1)+εt This assumes specific
restrictions on parameter relationships.
2. ADL (1,1) Model: Yt=λ1+λ2Yt−1+λ3Xt+λ4Xt−1+εt This is more general and includes
lagged terms without restrictions.

Testing Procedure:
• A common factor test checks whether the restrictions imposed by the AR (1) model
are valid.
• If restrictions are rejected, it suggests that the ADL (1,1) model provides a better fit.

Results:
• The unrestricted ADL (1,1) model was estimated, and its residual sum of squares
(RSSU=0.000906) was compared to that of the restricted AR (1) model
(RSSR=0.001360).
• The test statistic was calculated as: n log (RSSR/RSSU) With n=35, this yielded a value
of 14.22, which exceeds the critical value at a 0.1% significance level (χ2=13.82).

You might also like