Empirical Methods in Finance Time Series Models Part 2: ARIMA Models by Sakshi Sharma

Empirical Methods In
Finance
Time series Models Part 2
ARIMA models
By
Sakshi Sharma
Why ARIMA or ARMA models
Stationarity?
Some examples
Autocorrelation
Stationarity and Correlogram in e
views
 How to test
 How to remove
Basics of ARIMA models
ARIMA methodolgy
 Box-Jenkins Methodology
Building ARMA Models
- The Box Jenkins Approach
 Box and Jenkins (1970) were the first to approach the task of
estimating an ARMA model in a systematic manner. There are
3 steps to their approach:
Step 1. Identification
Step 2. Estimation
Step 3. Model diagnostic checking and forecasting

 Step 1:
- Involves determining the order of the model.
- Use of graphical procedures
- A better procedure is now available
Building ARMA Models
- The Box Jenkins Approach (cont’d)
 Step 2:
- Estimation of the parameters
- Can be done using least squares or maximum likelihood
depending on the model.
 Step 3:
- Model checking
 Box and Jenkins suggest
- deliberate overfitting
- residual diagnostics
ARIMA Modelling
 The “I” stands for integrated – an integrated process is

one that is stationary.
 Typically financial researchers difference the variable
so that it is stationary
 Δyt = yt – yt-1 - differenced once
 ΔΔyt = Δyt – Δyt-1 - differenced twice
 ARIMA(p, d, q,) - d denotes the number of times the

variable is differenced (typically only once, occasionally
twice)
 An ARMA(p,q) model in the variable differenced d times is
equivalent to an ARIMA(p,d,q) model on the original data
Identification process
Identification
Identification
ACF and PACF
RAW GDP Correlogram
Estimation
Some More Recent Developments in
ARMA Modelling
• Identification would typically not be done using ACF’s.
• We want to form a parsimonious model.
• Reasons:
- variance (hence S.E.) of estimators is inversely proportional
to the number of degrees of freedom
• This gives motivation for using information criteria, which

embody 2 factors
- a term which is a function of the RSS
- some penalty for adding extra parameters
• The objective is to choose the number of parameters which
minimises the information criterion.
Information Criteria for Model Selection
 Embodying two factors: (i) a function of the residual sum
of squares (RSS) and (ii) a penalty for the loss of
degrees of freedom from having too many parameters
 Akaike (AIC); Schwarz Bayesian (SBIC); Hannan-Quinn
(HQIC)
2k
AIC  ln(ˆ 2 ) 
T
k
SBIC  ln(ˆ 2 )  lnT
T
2k
HQIC  ln(ˆ 2 )  ln(ln(T ))
T
2
where 
ˆ  RSS /T  k
T is the number of observations, k is the number of

parameters, which for an ARMA model with a constant is
equal to p+q+1.
Which should you use? No criterion is superior to the others
– see Chris Brooks p. 257-8 for further discussion.
Estimation
Diagnostics ( contd…)
3. Diagnostics
Restimate
3. Diagnostics
Forecasting
Forecasting in Econometrics
 Determining the value that a variable is likely to take

 Why forecast?
 to predict returns to holding an asset
 to predict the riskiness of a portfolio
 to predict correlation between two variables
 Econometric (structural) forecasting:
 relies on relating a dependent variable to movements
in independent variables
 Time series forecasting:
 - relies on predicting future values on the basis of
past values
Forecasting in Econometrics
 Point forecasts : a single value prediction
 Interval forecasts: an indication of the range within which
a value is likely to lie
 In-sample forecast: retaining observations from a time-
series to investigate model reliability
 Out-of-sample forecast: predictions of future values not
yet observed
 The forecasting horizon:
 - one-step-ahead: forecast for the next observation only
 multi-step-ahead: forecast for 1,2,3,…,s observations
ahead
 Forecasts become less reliable as the forecast horizon is
extended.
In-Sample Versus Out-of-Sample
 Say we have some data - e.g. monthly FTSE returns for 120
months: 1990M1 – 1999M12. We could use all of it to build the
model, or keep some observations back:
 A good test of the model since we have not used the

information from 1999M1 onwards when we estimated the
model parameters
Because multi-step-ahead forecasts become less reliable
further ahead, one solution is to use a rolling or recursive
window for estimation:
Objective: To produce Data Used for Estimation of Parameters
3 step ahead forecasts Rolling Window Recursive Window

2002 M1, M2, M3 1990M1 – 2001M12 1990M1 – 2001M12
2002 M2, M3, M4 1990M2 – 2002M1 1990M1 – 2002M1
2002 M3, M4, M5 1990M3 – 2002M2 1990M1 – 2002M1
Etc.
The sample size for the rolling estimation window remains

fixed.
The sample size for the recursive window increases by 1

each step.
Determining the Accuracy of a Forecast
Forecast error = Actual – Forecast
1 T
Mean Squared Error: MSE   (y t  s  f t ,s ) 2
T  (T1  1) t T1
where T is the total sample size (in- and out-of-) and T1 is
the first out-of-sample forecast observation.
1 T
Mean Absolute Error: MAE   | y t  s  f t ,s |
T  (T1  1) t T1
Mean Absolute Percentage Error:

100 T y  ft s
t  s
MAPE  
T  (T1  1) t T1 y t s
has the advantage that it can be interpreted easily as it lies
between 0 and 100, but cannot be used if forecasts can be
both +ve and –ve (share returns for example).

Empirical Methods in Finance Time Series Models Part 2: ARIMA Models by Sakshi Sharma

Uploaded by

Copyright:

Available Formats

Empirical Methods in Finance Time Series Models Part 2: ARIMA Models by Sakshi Sharma

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Empirical Methods in Finance Time Series Models Part 2: ARIMA Models by Sakshi Sharma

Uploaded by

Copyright:

Available Formats

Empirical Methods In

 The “I” stands for integrated – an integrated process is

 ARIMA(p, d, q,) - d denotes the number of times the

• This gives motivation for using information criteria, which

T is the number of observations, k is the number of

 Determining the value that a variable is likely to take

 A good test of the model since we have not used the

Objective: To produce Data Used for Estimation of Parameters

3 step ahead forecasts Rolling Window Recursive Window

2002 M2, M3, M4 1990M2 – 2002M1 1990M1 – 2002M1

2002 M3, M4, M5 1990M3 – 2002M2 1990M1 – 2002M1

The sample size for the rolling estimation window remains

The sample size for the recursive window increases by 1

Mean Absolute Percentage Error:

You might also like