Financial Econometrics
Coursework assignment

According to the Purchasing Power Parity theory of nominal exchange rate determination, at time t a particular bundle of goods should cost exactly the same either:

(i) if it is purchased in the UK for a given price in £, say = £100; or
(ii) if it is purchased in the US from the proceeds of converting £100 into $ at the current nominal exchange rate.

If the purchase price in the US is = $150, PPP implies that the nominal exchange rate should be St = 1.5, i.e. £1 = $1.50.

An implication of PPP is that if price inflation is running at different rates in the two countries, the nominal exchange rate should adjust so that the PPP condition is maintained.

For example, suppose UK price inflation is 10% between year t and year t+1, and US price inflation is 5%.

In the UK, the bundle of goods will cost = £100×1.1 = £110
In the US, the bundle of goods will cost = $150×1.05 = $157.5

Therefore the nominal exchange rate required for the PPP condition to be maintained is St+1 = 1.4318 (=1.50×1.05/1.1). i.e. £1 = $1.4318. The £ has depreciated in value against the $.

An alternative way of expressing the PPP condition is to say that the real exchange rate, defined as , should always be constant.

Applying a log transformation,

where Qt = ln(Qt), , , st = ln(St)

Many empirical tests for the validity of the PPP theory of exchange rate determination have focused on:

either the stationarity/non-stationarity of qt (PPP ? qt should be stationary),

or the existence/non-existence of a cointegrating relationship between , and st (PPP ? , and st should be cointegrated).
The Excel file realexch.xlsx contains yearly time-series data for the period 1942-2014 for the following series:

puk = UK consumer/retail price index series,
pus = US consumer/retail price index series,
s0 = nominal $/£ exchange rate (US dollars per GBP), St

After loading the data into Stata, generate the natural logarithms of the three series (denoted below using lower-case symbols). Generate the real exchange rate series, and its natural logarithm. Suggested Stata variable names for the log series are lpuk, lpus, ls and lq.
1. Test each of the following series for stationarity or non-stationarity using the 68 observations for 1947-2014 only, using a Dickey-Fuller or Augmented Dickey-Fuller unit root test:

(i) (ii) (iii) st (iv) qt

In each case, use the Akaike Information Criterion to select the appropriate order (lag-length) for the DF/ADF(p) test, starting from p=4 and reducing p in steps of one as far as possible. Observations from years before 1947 can be used to create any lagged variables used in the DF/ADF autoregressions, but all DF/ADF autoregressions should be estimated over the 68 observations for 1947-2014 only.

For any series of the series that you find to be non-stationary, determine the order of integration by repeating the unit root test on the first-differences of the same series, and (if necessary) the second-differences.

Comment on the implications of (iv) for the validity of the PPP theory.

The tests completed in Q1 may produce evidence to suggest that one or more of , and st is I(2). In the following questions, however, for simplicity and for consistency with the PPP theory, we will assume that all three of these series have the same order of integration I(1).
2. Estimate a VAR model for ( , , ?st ) using the 68 observations for 1947-2014 only.

Use the multivariate Akaike Information Criterion to select the appropriate order (lag-length) for the VAR(p) model, starting from p=4 and reducing p in steps of one as far as possible.

Using your chosen model specification, carry out Granger causality tests of the following null hypotheses:

(i) Lagged values of and ?st do not Granger cause current values of .
(ii) Lagged values of and ?st do not Granger cause current values of .
(iii) Lagged values of and do not Granger cause current values of ?st.
3. Test for the existence of a cointegrating relationship between , and st using the Engle-Granger two-step residuals-based procedure:

Obtain the estimated cointegrating regression: using observations 1942-2014.

Save the residuals , and test for stationarity using the Engle-Granger adaptation of the ADF test, using the observations 1947-2014 only. Observations from years before 1947 can be used to create any lagged residuals used in the DF/ADF autoregressions, but the DF/ADF autoregressions should be estimated over the 68 observations for 1947-2014 only.

Comment on the implications of this cointegration test for the validity of the PPP theory.
4. Test for the existence of a cointegrating relationship between , and st using the Johansen Vector Error Correction Model (VECM) procedure:

Use the multivariate Akaike Information Criterion to select the appropriate order (lag-length) for the three-variable VAR(p) model for { , , st }, starting from p=4 and reducing p in steps of one as far as possible. Use the observations 1947-2014 only. Select the lag-length, p*, that produces the smallest value of MAIC.

Compute the Johansen rank and maximal eigenvalue cointegration tests based on the VECM derived from the VAR(p*) representation. Comment on the implications of this cointegration test for the validity of the PPP theory.

Estimate the VECM based on the VAR(p*) model with one cointegrating vector. Are the signs and statistical significance of the coefficients in the cointegrating vector consistent with the PPP theory?
General instructions

Your submission should comprise a written report, containing all relevant estimation and hypothesis test results presented in report form (as if you were writing for a journal article), and a brief commentary outlining the estimations and tests, and commenting on the results. There is no specific word limit for the commentary; a few sentences for each question will be sufficient. Marks will be awarded for clarity and presentation. You are encouraged to devise informative means for presenting the results, including well-designed tables. Your Stata output should be presented in an Appendix to the written report.

You may collaborate in working out how to run the estimations on Stata. However, you are required to write up your report independently, and I will require a separate and independent submission from each person.

Evaluating Regional Capital Mobility in China In the matlab .mat file, we have annual data on real per-capita consumption and real per-capital net income for nine regions in China. Between 1978 and 2007 (30 annual observations across 9 regions, so 260 pooled observations): Grouping of 21 provinces into 9 regions Region number Provinces included 1 Beijing, Tianjin, Hebei 2 Shanxi, Shandong 3 Liaoning, Jilin, Heilongjiang 4 Shanghai, Jiangsu, Zhejiang 5 Hunan 6 Henan, Anhui 7 Guangdong, Fujian 8 Guangxi, Yunan 9 Shaanxi, Xinjiang, Inner Mongolia The goal of the exercise is to evaluate the relationship between consumption and net income in these regions. Has it changed over time? If a country is becoming for integrated financially, it can borrow or lend within the broader national region, so that the link between consumption and net income in different regions should be less and less. Rich regions can lend to poorer regions and poorer regions can borrow from more prosperous ones. The measure of financial integration comes from the Feldstein-Horioka paradox. If there is free mobility of capital across nations or regions, then the link between national saving and national investment should be small. After all, under financial autarky, all investment has to be financed by domestic saving. Low and behold Professors Feldstein and Horioka discovered in the world economy, that there was a high correlation between national saving and national investment. This study challenged the prevailing rule that the capital markets are wide open and well integrated. Criticism of the Feldstein-Horioka result came from the fact that saving is measured differently across countries. So another approach was used: to test the relation between saving and investment across states, provinces or regions of particular countries. This is what we are going to do for China. You have a spreadsheet giving the real C and net income levels for the nine regions, in mcnelis_panel_finecon2015.xls. You are to estimate the model with a pooled sample and well as with fixed and random effects. You can also break the panel in half to see if there is a change in the more recent periods between consumption and net income. Word of caution: the data are in levels; we will have to put them in log first differences, to clear up auto correlation. There is also a likely relation between net income and the disturbance terms so you will likely want to use instrumental variables. Write up a three page essay telling me what you find in the data set: is there evidence of increasing financial integration in China? Attached is the word document and the xls sheet with data on real consumption and real net income (per capita) for nine regions of China. Your assignment is to examine these data to see if there is evidence of financial integration within China. That would imply a decreasing link within each region between Consumptions and net income, since borrowing and lending opportunities would be available through national capital markets. Attached is the spreadsheet with the panel data for the 9 regions of China, as well as a sample code for doing estimation with the data, for Panel data.. Panel data is really elementary. When we have data scross 9 regions and across time, we can either treat each region as different, with 9 different regressions, or we can pool all data into one sample, treat them as if they were all the same. Panel data is more realisitic: we are not completely different but we are also not completely identical. This is why we used fixed effects: across the regions some coefficients are the same, but some differ. Anyway for your extra credit, simply make use of this data set and tell me something interesting. You need the LeSage toolbox to do the regression. Note: for Mac users, use the matlab program Mypanel_China1.m and download the CHINA_DATA.mat file. The xlsread command does not work the same way on Mac versions of matlab. The m file should be run in the same directory as the speadhseet.

Evaluating Regional Capital Mobility in China In the matlab .mat file, we have annual data on real per-capita consumption and real per-capital net income for nine regions in China. Between 1978 and 2007 (30 annual observations across 9 regions, so 260 pooled observations): Grouping of 21 provinces into 9 regions Region number Provinces included 1 Beijing, Tianjin, Hebei 2 Shanxi, Shandong 3 Liaoning, Jilin, Heilongjiang 4 Shanghai, Jiangsu, Zhejiang 5 Hunan 6 Henan, Anhui 7 Guangdong, Fujian 8 Guangxi, Yunan 9 Shaanxi, Xinjiang, Inner Mongolia The goal of the exercise is to evaluate the relationship between consumption and net income in these regions. Has it changed over time? If a country is becoming for integrated financially, it can borrow or lend within the broader national region, so that the link between consumption and net income in different regions should be less and less. Rich regions can lend to poorer regions and poorer regions can borrow from more prosperous ones. The measure of financial integration comes from the Feldstein-Horioka paradox. If there is free mobility of capital across nations or regions, then the link between national saving and national investment should be small. After all, under financial autarky, all investment has to be financed by domestic saving. Low and behold Professors Feldstein and Horioka discovered in the world economy, that there was a high correlation between national saving and national investment. This study challenged the prevailing rule that the capital markets are wide open and well integrated. Criticism of the Feldstein-Horioka result came from the fact that saving is measured differently across countries. So another approach was used: to test the relation between saving and investment across states, provinces or regions of particular countries. This is what we are going to do for China. You have a spreadsheet giving the real C and net income levels for the nine regions, in mcnelis_panel_finecon2015.xls. You are to estimate the model with a pooled sample and well as with fixed and random effects. You can also break the panel in half to see if there is a change in the more recent periods between consumption and net income. Word of caution: the data are in levels; we will have to put them in log first differences, to clear up auto correlation. There is also a likely relation between net income and the disturbance terms so you will likely want to use instrumental variables. Write up a three page essay telling me what you find in the data set: is there evidence of increasing financial integration in China? Attached is the word document and the xls sheet with data on real consumption and real net income (per capita) for nine regions of China. Your assignment is to examine these data to see if there is evidence of financial integration within China. That would imply a decreasing link within each region between Consumptions and net income, since borrowing and lending opportunities would be available through national capital markets. Attached is the spreadsheet with the panel data for the 9 regions of China, as well as a sample code for doing estimation with the data, for Panel data.. Panel data is really elementary. When we have data scross 9 regions and across time, we can either treat each region as different, with 9 different regressions, or we can pool all data into one sample, treat them as if they were all the same. Panel data is more realisitic: we are not completely different but we are also not completely identical. This is why we used fixed effects: across the regions some coefficients are the same, but some differ. Anyway for your extra credit, simply make use of this data set and tell me something interesting. You need the LeSage toolbox to do the regression. Note: for Mac users, use the matlab program Mypanel_China1.m and download the CHINA_DATA.mat file. The xlsread command does not work the same way on Mac versions of matlab. The m file should be run in the same directory as the speadhseet.