Econometrics Board Questions:

1. CLRM Assumptions
CLRM stands for the Classical Linear Regression Model. It is a statistical framework used to
analyze the relationship between a dependent variable and one or more independent variables.
The CLRM assumes a linear relationship between the variables and employs ordinary least
squares (OLS) estimation to estimate the coefficients of the model.
The Classical Linear Regression Model (CLRM) makes several key assumptions to ensure the
validity of the ordinary least squares (OLS) estimation and statistical inferences. These
assumptions are as follows:

Linearity: The relationship between the dependent variable and the independent variables is
linear in nature. The model assumes that the true relationship can be adequately approximated by
a linear equation.

Independence: The observations in the dataset are independent of each other. This assumption
implies that the error terms or residuals are not correlated with each other. Independence ensures
that each observation provides unique information to the model.

Homoscedasticity: The error terms have constant variance across all levels of the independent
variables. In other words, the variability of the errors is the same for all predicted values of the
dependent variable. Homoscedasticity ensures that the OLS estimates are efficient and reliable.

No perfect multicollinearity: There is no perfect linear relationship between the independent

variables. Perfect multicollinearity occurs when one independent variable is a perfect linear
combination of other variables in the model. It can lead to issues in estimating the coefficients
and their interpretations.

Zero conditional mean: The expected value of the error term given the values of the
independent variables is zero. This assumption is also known as the exogeneity assumption. It
implies that the independent variables are not systematically related to the error term, ensuring
that the estimated coefficients are unbiased and consistent.

Normality: The error terms follow a normal distribution with a mean of zero. This assumption
allows for the application of inferential statistics and hypothesis testing. The normality
assumption is particularly important for small sample sizes as it enables valid statistical
inferences.

1
Violation of these assumptions can affect the accuracy and reliability of the OLS estimates and
statistical inferences. Diagnostic tests, such as residual analysis, tests for heteroscedasticity, and
tests for normality, can help assess the validity of these assumptions.

Multicollinearity

Multicollinearity refers to a situation in which two or more independent variables in a statistical

model are highly correlated with each other. It is a phenomenon where there is a strong linear
relationship between the predictor variables, making it difficult to determine their individual
effects on the dependent variable.
Multicollinearity, the presence of high correlation or linear dependency among independent variables in a
statistical model, can have several consequences. Here are some key consequences of multicollinearity:

Multicollinearity, or high correlation between independent variables, can have the following
consequences:
Unreliable estimates: It becomes difficult to accurately estimate the effects of each independent
variable on the outcome variable.
Uncertain variable importance: It is challenging to determine which variables are truly
important in predicting the outcome.
Confusing interpretations: The relationship between variables can be confusing, leading to
unclear interpretations of their impact on the outcome.
Difficult hypothesis testing: It becomes hard to determine if individual variables have a
significant impact on the outcome.
Reduced predictive accuracy: The model's ability to make accurate predictions decreases,
making it less reliable.
Detecting multicollinearity, or high correlation between independent variables, can be done
using the following simple methods:
Correlation matrix: Calculate the correlation coefficients between all pairs of independent
variables. If the correlation coefficients are close to 1 or -1, it indicates a high degree of
correlation and suggests multicollinearity.
Variance Inflation Factor (VIF): Calculate the VIF for each independent variable. A VIF value
greater than 1 suggests the presence of multicollinearity, and higher VIF values indicate stronger
multicollinearity.

2
Tolerance: Calculate the tolerance for each independent variable. A tolerance value less than 0.1
suggests the presence of multicollinearity.
Scatterplots: Plot the independent variables against each other to visually inspect their
relationship. If there is a strong linear pattern or clustering of data points, it suggests
multicollinearity.

When multicollinearity is detected, there are several remedies that can be applied to address the
issue. Here are some simple remedies for multicollinearity:
Remove correlated variables: If two or more variables are highly correlated, consider removing
one of them from the model. By eliminating one of the variables, you can reduce the
multicollinearity.
Transform variables: Instead of using the original variables, you can transform them into new
variables. For example, you can calculate the percentage change or take the logarithm of the
variables. Transformations can help reduce the correlation between variables.
Combine variables: If multiple variables are measuring similar aspects or have high correlation,
consider creating a composite variable by averaging or summing the correlated variables. This
reduces multicollinearity by consolidating the information from multiple variables into a single
variable.
Use regularization techniques: Regularization methods, such as ridge regression or lasso
regression, can help mitigate multicollinearity. These techniques introduce a penalty term that
reduces the impact of correlated variables and stabilizes the coefficient estimates.
Increase sample size: Multicollinearity can be more problematic with small sample sizes.
Increasing the sample size can help reduce the impact of multicollinearity and improve the
stability of coefficient estimates.

Heteroscedasticity
Heteroscedasticity, in simple terms, refers to a situation where the variability or spread of data
points is not the same across different values of an independent variable in a regression analysis.
It means that the scatter of points around the regression line changes as you move along the
independent variable.
Imagine plotting the points of a scatterplot with a regression line. If the points are spread out
more widely in some areas and closer together in others, it suggests heteroscedasticity. This
pattern indicates that the variability of the data points is not consistent throughout the range of
the independent variable.
3
Heteroscedasticity, or uneven variability of data points in a regression analysis, can have the
following consequences:
Biased coefficient estimates: Heteroscedasticity can lead to biased estimates of the relationships
between independent variables and the dependent variable. The coefficients may not accurately
represent the true effects of the independent variables.
Inaccurate standard errors: Heteroscedasticity can affect the calculation of standard errors,
which are used to determine the precision of the estimated coefficients. Incorrect standard errors
can result in unreliable hypothesis tests and confidence intervals.
Invalid hypothesis tests: Violation of the assumption of constant variance can make hypothesis
tests unreliable. Test statistics such as t-tests and F-tests may produce incorrect results, leading to
incorrect conclusions about the significance of the independent variables.
Inefficient predictions: Heteroscedasticity can affect the accuracy of predictions made by the
regression model. The model may perform well in some regions of the data but poorly in others,
as it fails to account for the varying variability of the data points.
Detecting heteroscedasticity, or uneven variability of data points in a regression analysis, can be
done using the following simple methods:
Scatterplot: Create a scatterplot of the residuals (the differences between the observed and
predicted values of the dependent variable) against the predicted values or the independent
variable(s). Look for a visible pattern in the spread of the residuals. If the spread widens or
narrows as the predicted values or independent variable(s) change, it suggests the presence of
heteroscedasticity.
Residual plot: Plot the residuals against the predicted values or the independent variable(s) and
examine the scatter of points. If the points show a funnel-like shape or exhibit a systematic
change in spread, it indicates heteroscedasticity.
Breusch-Pagan test: This statistical test formally examines the presence of heteroscedasticity in
a regression model. It assesses whether there is a significant relationship between the squared
residuals and the independent variables. If the p-value of the test is below a chosen significance
level (e.g., 0.05), it suggests the presence of heteroscedasticity.
White test: Similar to the Breusch-Pagan test, the White test is another statistical test for
heteroscedasticity. It examines whether the squared residuals are correlated with the independent
variables. A significant p-value indicates the presence of heteroscedasticity.
When heteroscedasticity is detected in a regression analysis, there are several simple remedies
that can be applied to address the issue:

4
Transform the variables: Apply transformations to the variables involved in the regression
model. Common transformations include taking the logarithm, square root, or inverse of the
variables. Transformations can help stabilize the variability of the data and mitigate
heteroscedasticity.
Weighted least squares regression: Give more weight to observations with smaller residuals or
lower variability. Weighted least squares regression assigns different weights to each observation
based on the estimated variance. This approach down weights observations with higher
variability, reducing the impact of heteroscedasticity on the estimates.
Robust standard errors: Calculate robust standard errors, also known as heteroscedasticity-
robust standard errors or White standard errors. These standard errors account for
heteroscedasticity and provide more reliable estimates of the coefficient standard errors,
hypothesis tests, and confidence intervals.
Non-linear regression: If the relationship between the variables suggests a non-linear pattern,
consider using non-linear regression techniques instead of linear regression. Non-linear
regression models can better capture the underlying structure and variability of the data.
Segment the data: If the heteroscedasticity is driven by certain groups or segments within the
data, consider analyzing each group separately or applying different models or techniques to
each segment. This approach allows for a more tailored analysis that accounts for the specific
heteroscedasticity patterns within each group.

Autocorrelation
Autocorrelation, in simple terms, refers to the correlation or relationship between observations of
a variable with its own past observations. It examines whether there is a systematic pattern or
dependence between the values of a variable at different time points.
Imagine you have a time series dataset where you measure a variable over multiple time periods.
Autocorrelation measures the similarity or relationship between the variable's value at a given
time point and its value at previous time points.
Positive autocorrelation occurs when high values of the variable tend to follow high values, or
low values tend to follow low values. This suggests a positive relationship or trend in the data.
Negative autocorrelation, on the other hand, occurs when high values of the variable tend to
follow low values, or low values tend to follow high values. This indicates a negative
relationship or trend in the data.
Autocorrelation, or the presence of a systematic relationship between a variable and its own past
values in a time series, can have the following consequences:
5
Biased coefficient estimates: Autocorrelation can lead to biased estimates of the relationships
between the independent variables and the dependent variable. The estimated coefficients may
not accurately represent the true effects of the variables.
Inefficient standard errors: Autocorrelation can affect the calculation of standard errors, which
are used to determine the precision of the estimated coefficients. Incorrect standard errors can
result in unreliable hypothesis tests and confidence intervals.
Invalid hypothesis tests: Violation of the assumption of no autocorrelation can make hypothesis
tests unreliable. Test statistics such as t-tests and F-tests may produce incorrect results, leading to
incorrect conclusions about the significance of the independent variables.
Inaccurate forecasts: Autocorrelation patterns can impact the accuracy of forecasts made using
time series models. If autocorrelation is not accounted for, the model may not capture the
underlying patterns in the data, leading to inaccurate predictions.
Serial correlation: Autocorrelation is also known as serial correlation. It implies that the error
terms in the model are correlated over time. This violates the assumption of independent and
identically distributed errors, which is crucial for reliable statistical inference.
Detecting autocorrelation, or the presence of a systematic relationship between a variable and its
own past values in a time series, can be done using the following simple methods:
Visual inspection: Plot the time series data and examine the pattern of the data points over time.
Look for any obvious trends or cycles that indicate a potential presence of autocorrelation.
Autocorrelation function (ACF) plot: Calculate the autocorrelation coefficients for different
lags and plot them on an ACF plot. The ACF plot shows the correlation between the variable and
its past values at various lags. If the autocorrelation coefficients exceed the confidence intervals,
it suggests the presence of autocorrelation.
Partial autocorrelation function (PACF) plot: The PACF plot shows the correlation between
the variable and its past values after removing the influence of intermediate lags. It helps identify
the direct influence of each lag on the current value. If the partial autocorrelation coefficients
exceed the confidence intervals, it indicates autocorrelation.
Durbin-Watson test: The Durbin-Watson test is a statistical test that checks for the presence of
autocorrelation in the residuals of a regression model. The test statistic ranges from 0 to 4, and
values close to 0 indicate positive autocorrelation, while values close to 4 suggest negative
autocorrelation. Values around 2 indicate no autocorrelation.
When autocorrelation is detected in a time series analysis, there are several simple remedies that
can be applied to address the issue:

6
Differencing: Take first differences or higher-order differences of the variable. This involves
subtracting each observation from its lagged value. Differencing can help remove autocorrelation
by eliminating the trend or serial dependence in the data.
Include lagged variables: Incorporate lagged values of the variable as additional independent
variables in the model. By including past values as predictors, you can capture the
autocorrelation patterns and reduce the autocorrelation in the residuals.
Autoregressive Integrated Moving Average (ARIMA) models: ARIMA models are
specifically designed to handle autocorrelation in time series data. These models incorporate
autoregressive (AR), differencing (I), and moving average (MA) components to capture the
autocorrelation patterns. ARIMA models can be estimated and used for forecasting.
Seasonal adjustments: If the data exhibits seasonal patterns, seasonal adjustments can be
applied to remove autocorrelation. Seasonal adjustments involve removing the seasonal
component from the data, typically using techniques such as seasonal decomposition of time
series or seasonal adjustment methods like seasonal ARIMA (SARIMA).
Transformations: Apply transformations to the data to stabilize the variance or reduce the
autocorrelation. Common transformations include logarithmic, square root, or Box-Cox
transformations. Transformations can help make the data more stationary and mitigate
autocorrelation.

Time Series Analysis and Forecasting

Stationary
In time series analysis, stationarity refers to a property of a time series where the statistical
properties of the data remain constant over time. It is an important assumption for many time
series models and statistical tests.
There are three components to consider when assessing stationarity in a time series:
Constant mean: The average value of the time series remains the same over time. There are no
significant upward or downward trends in the data.
Constant variance: The variability or spread of the data points remains consistent over time. The
distribution of the data does not widen or narrow as time progresses.Constant autocovariance or
autocorrelation: The correlation between the values of the time series at different time points
remains constant. The relationship between the current observation and past observations does
not change over time.

7
A stationary time series allows us to make reliable predictions and draw meaningful conclusions
from the data. It simplifies the modeling process because the statistical properties of the data are
stable.
If a time series is not stationary, it can be transformed to achieve stationarity. Common
techniques include differencing, where the difference between consecutive observations is taken,
and seasonal adjustment, which removes seasonality effects. These transformations aim to
remove trends, seasonality, or other patterns that cause non-stationarity.
It's worth noting that stationarity is a property of the time series itself and not necessarily the
individual data points. Even if the data points exhibit some variation, the overall time series can
still be stationary if the statistical properties remain constant.
In summary, stationarity in a time series means that the mean, variance, and autocorrelation
structure of the data remain constant over time. Achieving stationarity is important for accurate
analysis and modeling of time series data.

Non-Stationary
In time series analysis, non-stationarity refers to a property of a time series where the statistical
properties of the data change over time. It means that the mean, variance, or autocorrelation
structure of the data exhibits some form of trend, seasonality, or other patterns that evolve over
time.
Non-stationary time series can have various characteristics:
Trend: The time series shows a systematic upward or downward movement over time. It
indicates a long-term increase or decrease in the data points.
Seasonality: The time series exhibits repetitive patterns or cycles at fixed intervals, such as daily,
weekly, or yearly patterns. Seasonality reflects consistent variations that repeat within a specific
time frame.
Changing variance: The spread or variability of the data points varies over time. The distribution
of the data widens or narrows as time progresses.
Time-dependent autocorrelation: The correlation between the values of the time series at
different time points changes over time. The relationship between the current observation and
past observations is not constant.
Non-stationarity poses challenges for time series analysis and modeling because it violates the
assumptions of many statistical techniques. Predictions and conclusions drawn from non-
stationary time series may not be reliable or meaningful.

8
To address non-stationarity, various techniques can be applied:
Differencing: Taking the difference between consecutive observations can help remove trends
and achieve stationarity.
Detrending: Removing the trend component from the time series can make it stationary. This can
be done using techniques like regression analysis or moving averages.
Seasonal adjustment: Removing the seasonal component from the time series can eliminate
seasonality effects. Techniques like seasonal decomposition of time series or seasonal adjustment
models can be used.
Transformation: Applying mathematical transformations, such as logarithmic or power
transformations, can stabilize the variance and make the series stationary.

ARIMA
ARIMA, or Autoregressive Integrated Moving Average, is a forecasting model used to predict
future values in a time series dataset. It combines three components: autoregressive (AR),
differencing (I), and moving average (MA).
Autoregressive (AR): This component looks at the relationship between an observation and a
certain number of its previous values. It assumes that the future value of a variable depends on its
past values.
Differencing (I): This component helps make the time series stationary by taking the difference
between consecutive observations. Stationarity means that the statistical properties of the data,
such as mean and variance, remain constant over time.
Moving Average (MA): This component takes into account the relationship between an
observation and the errors or residuals from past forecasts. It helps capture short-term
fluctuations or random shocks in the time series.
The ARIMA model is denoted as ARIMA (p, d, q), where:
p represents the order of the autoregressive component (number of past values considered).
d represents the order of differencing required to achieve stationarity.
q represents the order of the moving average component (number of past forecast errors
considered).
By estimating the values of p, d, and q based on the characteristics of the time series data, an
ARIMA model can be used to make predictions about future values in the series.

9
The Box-Jenkins methodology
The Box-Jenkins methodology, also known as the Box-Jenkins approach or Box-Jenkins’s
method, is a systematic and iterative approach to time series analysis and forecasting. It was
developed by George Box and Gwilym Jenkins and has become a widely used framework for
modeling and forecasting time series data.
The Box-Jenkins methodology is a step-by-step approach used to analyze and forecast time
series data. Here's a simplified explanation of the methodology:
Identify patterns: First, we examine the time series data to find any patterns, trends, or seasonal
variations. This helps us understand the underlying structure of the data.
Select a model: Based on the identified patterns, we choose an appropriate ARIMA model. This
model consists of autoregressive (AR), moving average (MA), and differencing (I) components.
The orders of these components are determined by analyzing autocorrelation and partial
autocorrelation plots.
Estimate parameters: Next, we estimate the parameters of the selected ARIMA model using
statistical methods. These parameters describe the relationships between past and current values
of the time series.
Check model fit: We evaluate how well the ARIMA model fits the data by examining the
residuals (the differences between the predicted and actual values). Diagnostic tests and plots are
used to check for any remaining patterns or problems in the model.
Refine the model: If the model doesn't fit well or if there are residual patterns, we refine the
model by adjusting the orders of the AR, MA, and differencing components. This iterative
process continues until a satisfactory model is obtained.
Forecast future values: Once we have a good model, we can use it to make predictions for
future values of the time series. The model's parameters and historical data are used to generate
forecasts.
Evaluate and update: We evaluate the accuracy of the forecasts by comparing them to the
actual values. If necessary, we update the model periodically with new data to improve its
forecasting performance.

Simultaneous equations:
Simultaneous equations refer to a situation where we have a set of equations where the variables
are interrelated and affect each other. Imagine a scenario where the value of one variable

10
depends on the values of other variables in the system. These equations need to be solved
together to find the values of all variables that satisfy all equations at the same time.
Each equation in the system represents a relationship between the variables. The solution to the
simultaneous equations is a set of values that makes all the equations true at the same time. The
number of equations should be equal to the number of unknown variables to have a unique
solution.
Simultaneous equations can be linear or nonlinear, depending on whether the equations involve
linear or nonlinear relationships between the variables. The goal is to find the values of the
variables that satisfy all the equations in the system.

Instrumental Variables (IV):

Instrumental variables (IV) is a technique used to address a common problem in economics
called endogeneity. Endogeneity occurs when there is a relationship between the variables we are
interested in and other factors that are not directly included in the equation but affect the
outcome. IV helps us establish a causal relationship between variables by introducing an external
variable, called an instrument.
The instrument is a variable that is related to the variable of interest but not affected by the
factors causing endogeneity. By using the instrument, we can estimate the causal relationship
between the variables accurately. The instrument acts as a substitute or proxy for the variable we
are interested in, allowing us to overcome the endogeneity problem and obtain reliable estimates.
In simple terms, simultaneous equations involve a system of equations where variables are
interrelated and affect each other. To solve these equations, we need to find values that satisfy all
equations at the same time. Instrumental variable (IV) is a technique used to address the problem
of variables being influenced by factors outside the equation. It introduces an external variable
that is related to the variable of interest but not affected by the outside factors, allowing us to
estimate the causal relationship accurately.

Logit and Probit

Logit and Probit are two statistical models used to analyze situations where we have two possible
outcomes, such as yes/no, success/failure, or buy/don't buy. These models help us understand
and predict the probability of one of these outcomes happening.
In both models, we have some factors or variables that we think might influence the outcome.
For example, in predicting whether someone will buy a product, these factors could include their
income, age, or gender.
The logit model uses a function called the logistic function to model the relationship between
these factors and the probability of the outcome. It gives us a value between 0 and 1,
11
representing the probability of the outcome happening. This model is commonly used in
economics and social sciences.
The probit model, on the other hand, uses a different function called the cumulative distribution
function of the standard normal distribution. It also provides a value between 0 and 1,
representing the probability of the outcome. This model is often used in fields like medicine and
psychology.
Both models estimate the influence of each factor on the outcome by finding the best fit to the
observed data. They give us coefficients that indicate the direction and strength of the
relationship between the factors and the outcome. These coefficients help us understand how
each factor affects the probability of the outcome.
To summarize, logit and probit models are used when we have a binary outcome and want to
understand how different factors influence the probability of that outcome. They use different
mathematical functions to model this relationship and provide coefficients that tell us the impact
of each factor.

What is co-integration? Explain in detail 

Co-integration is a valuable technique used to identify the presence of a long-term relationship
among two or more non-stationary variables. Its purpose is to uncover how these variables move
in sync over time. The significance of co-integration lies in its ability to prevent spurious
regression, a pitfall that occurs when non-stationary variables are erroneously regressed against
each other. 
To examine co-integration, we can employ the Engle-Granger two-step method. The initial step
involves testing the variables for the order of integration. If they share the same integration
order, we can utilize ordinary least squares (OLS) to estimate the long-run equilibrium model
and save the residuals. In the subsequent step, the residuals are assessed for stationarity. If they
exhibit stationarity, we can conclude that the variables are indeed co-integrated. 
Once co-integration is established, we can proceed to estimate the error correction model (ECM),
also known as the short-run model. The ECM enables us to comprehend the speed at which the
variables adjust towards their long-term equilibrium relationship. It incorporates lagged values of
the variables and the error correction term (ECT), which represents the disparity between the
actual and predicted values of the dependent variable.  
 Variables: Export (x) – Imports (M) for Canada. 1960Q1 – 2005Q5 
Long run model  
𝑥𝑡=−0.45+1.02𝑚𝑡+𝜀𝑡xt=−0.45+1.02mt+𝜀t
  
Short run model  

12
 𝑥𝑡=0.009+0.64𝑚𝑡+−011𝜀𝑡−1+𝑣𝑡xt=0.009+0.64mt+−011𝜀t−1+vt
  
Model 
variable  
Long Run  P-value  Short Run  P-value 
C  -0.451  0.004  0.0009  0.000 
M  1.02  0.000  0.638  0.000 
ECT        -0.116  0.0006 
 
 The ECT is statistically significant. 
 The coefficient of the ECT is -0.1165, suggesting that almost 12% the
discrepancy between the long run and the short run is corrected within a quarter. 
 Explain ourselves 

13