EC50162 Coursework
Practical Assignment (30%)
Please write up a short report based on the following exercise. The report should be a maximum of 2000 words and should include answers to all parts of the questions. Please include all relevant computer printouts in the Appendix.
Deadline: Friday May 12th 2006 (by 12.00)
Using the data on the file FTDATA, you will be allocated a company’s share return (To follow) to conduct the following tests. The file also contains ‘MRPI’, the return to the market and TB1 a treasury bill interest rate. After creating a constant, carry out the following (Don’t forget a negative result is just as interesting as a positive result, you will be marked on how well you conduct these tests and interpret the results, not on how good the results are).
1)Carry out tests for stationarity on your company (i.e. r1), MRPI and TB1. Firstly by plotting the data, then through assessing an appropriate correlogram and Llung-Box statistic, finally by Augmented-Dickey Fuller (ADF) tests. In each case only stipulate 4 lags, asset markets adjust quickly, so a long lag structure is not necessary. Interpret the results, based on your output here, you need to decide on the number of times you need to first-difference your variables to produce stationarity. (Don’t forget, it could be they don’t need first-differencing!)
2)With your company return, assess how well it performs as an AR(1) process. (i.e. the return regressed on a constant and lagged return). Assess the diagnostic tests and explanatory power. Add the monthly dummy variables contained in the data file (jan, feb etc), does this improve the result? If not remove them.
3)Does your company perform better if you include a moving average process? Try an ARMA (1,1) process.
4)How well does your model perform if you include 4 lags in the AR and MA process? Either with reference to the Akaike criteria, Schwarz-Bayesian criteria or statistic, find the optimal selection of lags for your ARMA model. Assess the model with respect to the diagnostic tests and explanatory power.
5)Try out-of-sample forecasting with your model from (4), leave 20 observations at the end of the sample for the out-of sample forecasting.
6)To test if the asset follows an ARCH type of process, run a simple AR(1) model for your return and save the residuals. Then square the residuals and carry out a secondary test. This involves the following regression:
Then calculate the statistic, which follows the Chi-squared (4) distribution. What are your conclusions (It is not necessary to produce an ARCH model)
7)Carry out a Vector Autoregressive (VAR) test on the following model:
Where r1 is the company share return allocated to you, select 4 as the lag length for your VAR. Assess the results and diagnostic tests, also carry out a forecast with this model and assess its performance, as in part (5).
8)Does a Vector Error Correction (VECM) improve the results?
9)Based on the results from above assess the overall performance of your company share return, how would you best model it?