Journal of International Money and Finance 19 (2000) 753757 www.elsevier.

nl/locate/econbase

Purchasing power parity over two centuries?

John T. Cuddington
a

a,*

, Hong Liang

Department of Economics, ICC 580, Georgetown University, Washington, DC 20057-1036, USA b The International Monetary Fund, 700 19th Street, NW, Washington, DC 20431, USA

Abstract This paper re-examines the purchasing power parity hypothesis for the dollarsterling exchange rate using the two centuries of data from Lothian and Taylor (LT) (1996) [Real exchange rate behavior: the recent oat from the perspective of the past two centuries. Journal of Political Economy 104 (3), 488509]. Unlike LT, we conclude that the dollarsterling RER is nonstationary, implying a rejection of the long-run PPP hypothesis. The differences in our conclusions are explained by: (1) sensitivity of ADF unit root tests to the choice of lag length, and/or (2) the presence of signicant time trends in the ADF or PhillipsPerron unit root test equations. 2000 Elsevier Science Ltd. All rights reserved.
JEL classication: F3 (international nance); F41 (open economy macroeconomics) Keywords: Purchasing power parity; Real exchange rates; Unit root tests; Sensitivity analysis

1. Introduction In the last 1015 years, a large literature has emerged on testing the long-run validity of purchasing power parity (PPP), or equivalently the stationarity of the real exchange rate (RER), using modern time-series econometrics techniques. (See Rogoff, 1996, for recent references.) Lothian and Taylor (LT) (1996) emphasize that low power in standard unit root tests, especially with short data spans, may have caused researchers to incorrectly conclude that the RER is nonstationary. They present new unit root test results for the francsterling and dollarsterling RERs using annual time series spanning two centuries. With the increased test power obtained by this large data sample, they are able to reject the unit root hypothesis using both aug* Corresponding author. Tel.: +1-202-687-6103; fax: +1-202-687-6102. E-mail addresses: cuddingj@georgetown.edu (J.T. Cuddington), hliang@imf.org (H. Liang).
0261-5606/00/$ - see front matter 2000 Elsevier Science Ltd. All rights reserved. PII: S 0 2 6 1 - 5 6 0 6 ( 0 0 ) 0 0 0 2 9 - 2

754

J.T. Cuddington, H. Liang / Journal of International Money and Finance 19 (2000) 753757

mented DickeyFuller (ADF) and PhillipsPerron (PP) tests. They therefore conclude that PPP is valid in the long run for the two bilateral exchange rates considered. Although we agree with LT that the francsterling exchange rate is stationary, our re-examination of the dollarsterling RER concludes that it is not stationary. The differences in our conclusions are explained by: (1) sensitivity of the ADF unit root tests to the choice of lag length, and/or (2) the presence of signicant time trends in the ADF or PP unit root test equations. We argue that, with suitably long lag lengths in the ADF equations, the unit root hypothesis is not rejected. If the lag length is shortened to that considered by LT, the ADF equations have signicant time trends. The time trend is also signicant in the PP test equations. Either unit roots or deterministic time trends, of course, imply nonstationarity, and hence rejection of the PPP hypothesis. 2. Unit root tests and stationarity Unit root testing is hazardous terrain. In general, the appropriate procedure is to use the general-to-specic (GTS) methodology by rst estimating the ADF or PP1 unit root testing equation including both an intercept and time trend: qta0a1tgqt1

i 1

biqtiet

(1)

where qt is the logarithm of the RER. The arguments in favor of beginning with the most general specication including the intercept and time trend is the usual one involving omitted variable bias versus loss of efciency caused by redundant regressors. A time trend must be included initially to allow for the possibility of a deterministic trend in the alternative hypothesis when the null hypothesis of a unit root is tested (Hamilton, 1994).2 It is now well-known that ADF unit root tests are often sensitive to the choice of the lag length p in (1). Various criteria have been proposed in the literature. Hall (1994) and Ng and Perron (1995) argue convincingly, based on Monte Carlo analysis, that a GTS method they propose for lag selection dominates the Akaike and Shwartz criteria.3
Phillips and Perron (1988) provide a generalization of the ADF test that sets p=0 in (1), but allows for a weaker set of assumptions concerning the error process. 2 If the null hypothesis of a unit root is not rejected using (1), the signicance of the trend and intercept can then be tested in turn to see if they can be omitted from the test equation, thereby increasing the power of the unit root tests. See Enders (1995) for a detailed discussion of the GTS methodology and the appropriate critical values for testing the signicance of the trend and intercept terms in the various specications (in his Table 4.1). 3 Their lag selection method starts with a large number of lags, with the square root of the sample size being a good rule of thumb. Examine the t-statistic on the last lag (which is asymptotically normal). If it is insignicant, drop the last lag and re-estimate the test Eq. (1). Continue dropping the last lag in the lag polynomial, one by one, until a signicant lag (at, say, the 95% level) is found. Stop at that point, leaving all shorter lags in the regression and examine the signicance of the ADF t-statistic using the appropriate DF distribution.
1

J.T. Cuddington, H. Liang / Journal of International Money and Finance 19 (2000) 753757

755

Table 1 presents ADF tests for lag lengths from 15 (roughly T1/2) down to 0 to show how the choice of lag length affects the signicance of the ADF statistic and the time trend. Suppose the GTS lag selection method is used; it chooses p=14. This is considerably larger than the 5 lag specication used in LT. With 14 lags of the dependent variable in (1), the unit root hypothesis cannot be rejected. This implies a rejection of the PPP hypothesis for the dollarsterling exchange rate.4 Suppose that 14 lags is for some a priori reason considered to be too long. If the 14th lag is ignored by starting the GTS procedure at p=13, the chosen lag length is eight. The ADF test statistic (ADF=4.10) now indicates rejection of the unit root hypothesis, as did the ve-lag specication used by LT. Note, however, that in the eight-lag specication or any other lag choice between 0 and 10 for that matter, the coefcient on time (t) is statistically signicant. For p=8, the point estimate is 0.00059, with an associated t-statistic of 3.47.5 The PP unit root test,
Table 1 Sensitivity analysis on unit root test and trend signicancea No. of lags 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 ADF statistic 2.21 2.05 2.52 2.75 3.10 3.47* 3.42* 4.10* 3.62* 3.87* 4.30* 4.91* 4.88* 4.95* 5.03* 4.36* t-statistic on trend 1.87 1.75 2.13 2.29 2.61 2.90 2.68* 3.47* 3.04* 3.24* 3.53* 3.75* 3.52* 3.45* 3.38* 2.61*

Truncation lag 5

PhillipsPerron statistic 4.61*

t-statistic on trend 2.62*

a Note that the critical value for the t-statistic on the trend is 2.79, given the presence of a unit root, and 1.96 otherwise. *Indicates signicance at the 5% level.

4 Subsequent tests on the trend and constant nd that neither is signicant. Hence, a more restricted specication of (1) without these two terms is used in obtaining the ADF statistic. The conclusion is unchanged. The ADF t-statistic on g is 0.58, which is far below the DickeyFullerMacKinnon critical value of 1.95. 5 Note that the t-statistic on the time trend is asymptotically normal when the unit root hypothesis is rejected.

756

J.T. Cuddington, H. Liang / Journal of International Money and Finance 19 (2000) 753757

reported at the bottom of Table 1, indicates rejection of the unit root hypothesis, as in LT. Again, however, the time trend is statistically signicant.6 In Cuddington and Liang (1998), we report similar ndings for the two subperiods (17911945, 19461990) considered by LT. We also estimate trend stationary and difference stationary models that nest our preferred specications and those of LT. The restrictions implied by their specication are then tested and rejected. Our preferred univariate models stand up well to theirs (and a simple random walk alternative) in out-of-sample forecasting exercises.

3. Concluding remarks The ndings regarding the presence of a unit root test are sensitive to the choice of lag length in the ADF test. Nevertheless, all of the specications in Table 1 (for both the ADF and PP tests) imply non-stationarity of the real exchange rate over the past 200 years, either because of the presence of a unit root or a deterministic trend. Hence, the PPP hypothesis is rejected for the dollarsterling exchange rate. Theory tells us that PPP need not hold even in the long run. Recall, e.g., the BalassaSamuelson model where the failure of long-run PPP is due to differential rates of productivity growth in the tradable and nontradables sectors. Empirical evidence also suggests persistent deviations from the law of one price for traded goods,7 which may upset PPP. Moreover, the short-run size and direction of PPP deviations also appear to be closely related to nominal exchange rate movements.8 It is, however, difcult to see why these factors would lead to a deterministic time trend in the RER. Hence, we prefer the stochastic trend/unit root model, which is the specication suggested when longer lag lengths are chosen. The long lag length is also consistent with Rogoffs (1996) observation about the extreme persistence in the time series behavior of RERs. There is an intriguing question from our re-examination of PPP using the LT data:. Why does PPP apparently hold for the francsterling but not the dollarsterling exchange rate? There are several possible explanations. One is that the geographic distance is greater between the US and Europe than between the UK and France, but that effective distance has shrunk over time due to improvements in transportation and communications technology. Another possibility is that nominal exchange rate variability was lower between the UK and France than between the US and Europe. Interestingly, Engel and Rogers (1995) nd that the law of one price holds more closely for country pairs within the same geographic region. Theories on pricing to market and nominal price stickiness have offered many interesting ideas as

6 7 8

This nding is robust for any choice of the truncation lag (014). See Knetter (1989), Feenstra and Kendall (1994) and Ghosh and Wolf (1994). See Giovannini (1988) and Engel (1993).

J.T. Cuddington, H. Liang / Journal of International Money and Finance 19 (2000) 753757

757

to why distance and nominal exchange variability may matter for long-run RER movements.9 A full explanation, nevertheless, awaits further research.

Acknowledgements The authors thank Jim Lothian for providing the data set used in this study and Zhongmin Wang for research assistance.

References
Cuddington, J.T., Liang, H., 1998. Re-examining the purchasing power parity hypothesis over two centuries. Georgetown University working paper no. 98-01 (available at www.georgetown.edu/ cuddington). Enders, W., 1995. Applied Econometric Time Series. John Wiley & Sons, Inc. Engel, C., Rogers, J., 1995. Regional patterns in the law of one price: the roles of geography vs. currencies. National Bureau of Economic Research working paper no. s5395. Engel, C., 1993. Is real exchange rate variability caused by relative price changes? An empirical investigation. Journal of Monetary Economics 32, 3550. Feenstra, R., Kendall, J., 1994. Pass-through of exchange rates and purchasing power parity. National Bureau of Economic Research working paper no. 4842. Ghosh, A., Wolf, H., 1994. Pricing in international markets: lessons from the economist. National Bureau of Economic Research working paper no. 4806. Giovannini, A., 1988. Exchange rates and traded goods prices. Journal of International Economics 24, 4568. Goldberg, P.K., Knetter, M.M., 1997. Goods prices and exchange rates: what have we learned? Journal of Economic Literature XXXV, 12431272. Hall, A., 1994. Testing for a unit root in time series with pretest data-based model selection. Journal of Business and Economic Statistics 12 (4), 461470. Hamilton, J., 1994. Time Series Analysis. Princeton University Press. Knetter, M., 1989. Price discrimination by U.S. and German exporters. American Economic Review 79, 198210. Lothian, J.R., Taylor, M.P., 1996. Real exchange rate behavior: the recent oat from the perspective of the past two centuries. Journal of Political Economy 104 (3), 488509. Ng, S., Perron, P., 1995. Unit root tests in ARMA models with data-dependent methods for the selection of the truncation lag. Journal American Statistical Association 90 (429), 268281. Phillips, P., Perron, P., 1988. Testing for a unit root in time series regression. Biometrica 75, 335346. Rogoff, K., 1996. The purchasing power parity puzzle. Journal of Economic Literature 34 (2), 647668.

9 See Goldberg and Knetter (1997) for an excellent survey on recent literature on goods prices and the nominal exchange rates.