RDP 9203: Real Exchange Rates and the Globalisation of Financial Markets 3. The Changing Degree of International Financial Market Integration

The removal of officially-imposed barriers to the international movement of capital commenced in the United States shortly after the breakdown of the Bretton Woods system of fixed exchange rates. The severity of both capital and exchange control barriers has been progressively diminished elsewhere since then, with the result that these controls have now been virtually eliminated in the major OECD countries[6]. This should have facilitated the integration of international financial markets. Financial innovation, spurred by tremendous advances in telecommunications technology, should also have contributed to the increasing pace of integration. The growing international availability of new financial instruments such as currency and interest rate swaps and financial futures and options has encouraged international portfolio diversification by providing a wider array of financial instruments than are likely to be available on domestic financial markets. The level of cross-country integration is also likely to have been further facilitated by internationalisation in the provision of financial services, with foreign-based financial intermediaries playing an increasingly important role in domestic banking and securities markets.

A variety of approaches has been suggested for quantifying the degree of international financial market integration. These different measures, presented in Table 3 in descending order of specificity, do not typically give the same impression of the degree of integration. There are a few reasons for this. Some measures are more narrowly based (1, 2 and 3) than others, in the sense that the array and maturity of the assets implicitly included is restricted. Also, some tests are based on nominal magnitudes (1, 2 and 3) while others concern real variables (4 and 5). Most are based on the co-movement of relative prices, but 5 is based on the absence of co-movement between domestic saving and investment. To obtain a more comprehensive perspective on the degree of international financial integration, how this is changing over time, and how it might be related to goods market integration, all the measures presented in Table 3 are reviewed.

(a) Closed, Covered and Uncovered Interest Parity

The first definition of financial market integration is closed interest parity. This says that capital flows equalise interest rates on comparable financial instruments issued in different countries but denominated in the same currency. Of the five definitions this is the least stringent in that, for it to be valid, it requires the least number of conditions to be fulfilled. However, it is also the most narrrowly based in that it refers only to that subset of assets traded in Eurodeposit or Eurobond markets. These constitute only a small proportion of the value of financial instruments issued on the domestic market. Thus a conclusion that closed interest parity is valid clearly does not permit one to infer that international financial markets are completely integrated, but simply that the markets under consideration are.

The only reason for deviations from closed interest parity is the existence of a political risk premium. This is interpreted very widely here as representing not only existing capital controls and asset tax arrangements in different political jurisdictions, but also the prospect that existing barriers and taxes will change in the future[7]. A higher interest rate locally than offshore indicates barriers preventing capital inflows, while a negative differential reflects barriers preventing capital flight (except in countries such as the US and Switzerland where “safe haven” factors may be important). Reduced political barriers to trade in assets between onshore and offshore financial centres, or the prospect of such a development, will manifest itself as smaller deviations from closed interest parity[8]. Chart 2 displays the differential between three-month onshore (interbank) and offshore (Eurodeposit interbank) rates for seven OECD countries over periods for which data were available[9]. Even for assets denominated in the same currency, interest rate differentials have been very large in the past but have now, to all intents and purposes, disappeared. In most cases the elimination of the differential dates from the moment when capital controls were finally removed.

The second definition of international financial market integration is covered interest parity. This relates to yields on comparable assets issued in different countries and denominated in different currencies, namely the currencies of the issuing countries. Therefore, in addition to political risk, there is also currency risk. But insurance can be bought against the latter by resorting to the forward foreign exchange market or, for longer-term maturity instruments, the swap market. The difference between foreign asset yields hedged in the forward market (to compensate for expected exchange rate changes) and domestic yields also constitutes a measure of “political” risk. Thus, to the extent that currency transactions costs in the forward foreign exchange market are relatively small, covered interest rate disparities between any two countries should display similar patterns over time to the difference between the closed interest rate disparities for each of the two countries[10]. Covered disparities between US and other countries' three-month interest rates are graphed in Chart 3. The graphs corroborate visually the hypothesis that deviations from covered interest parity between national markets have declined substantially in recent times. While convergence to zero is not guaranteed because of minor conceptual differences in the data used, volatility in yield differentials has been reduced for most currencies (although not nearly as dramatically as in the case of closed interest parity).

The third measure set out in Table 3 is uncovered interest parity, which requires the domestic interest rate to equal the foreign interest rate on a comparable asset plus the expected depreciation of the domestic currency over the period to the maturity of the asset. If covered interest parity is assumed to be valid, a weak assumption especially for the Eurocurrency market, tests of uncovered interest parity are essentially tests of the efficiency of the market for foreign exchange[11]. The null hypothesis of uncovered interest parity is unanimously rejected virtually without exception. The agnostic inference following from this is that either expectations are not rational (there are systematic errors in forecasting exchange rates over the period concerned), a time-varying risk premium exists, or both conditions prevail. Recent evidence (e.g. Frankel and Froot (1987), Froot and Frankel (1989)) tends to suggest that inefficient expectational episodes dominate time-varying risk premia as explanations in this respect. Indeed the rejection of the null hypothesis is also consistent with a host of other phenomena such as bubbles, bandwagon and peso effects. The failure of uncovered interest parity to hold is, therefore, perfectly consistent with the removal of all administrative barriers to the free flow of capital – it does not undermine the assumption of increasing integration of world financial markets.

(b) Real Interest Parity

It is natural to presume that international investors are concerned with the expected purchasing power of the return on their investments, domestic and foreign, rather than just the nominal returns. If the exchange rate is expected to move to eliminate discrepancies between expected national inflation rates over the relevant asset holding period, then real interest rates may be the relevant relative price determining capital flows (as in the above model). It is arguable, accordingly, that real interest parity may be the appropriate criterion of international financial market integration. For real interest parity to hold it is necessary for ex ante real rates to equal, or to move to equality rapidly after a disturbance. This requires both uncovered interest parity and ex ante PPP to be valid (see Annex for an algebraic treatment). The evidence concerning short-run real interest rates, for which precise ex ante measures can be constructed is reviewed first. Subsequently longer-term real rates are considered.

Results of econometric tests for the co-movement of short-term real rates with those of the United States are reported in Table 4 and with those of Germany (for European countries) in Table 5. The co-movement of other countries' real short rates with that of the larger country is reflected in the size and significance of the estimated γ coefficients:

where expected real interest rates refer to short rates. The Annex describes how precise ex ante real rates are calculated. In this framework α=0, γ=1.0 implies complete equality of rates; α≠0, γ=1.0 implies integration since real rates move together; 0<γ<1.0 implies lack of complete integration; and γ=0 implies zero intergration.

The equations for rates vis-a-vis the United States were estimated for three time periods. The first two periods (August 1974 to October 1979) and (November 1979 to February 1990) were chosen in an attempt to identify whether the ongoing process of financial liberalisation and innovation altered the nature of the real interest parity relationship. The third period (January 1986 to February 1990) was chosen to see if the estimated results are robust to the sample selection involved in choosing October 1979 as the important breakpoint. If the closeness of the co-movement of real interest rates across countries was a reliable measure of the degree of international financial market integration, then one would expect to observe a stronger relationship and higher values of γ in the second compared to the first period.

The hypothesis of zero linkage between real rates in the United States and those in Germany is accepted in the first sub-period[12]. The hypothesis of only partial linkage is accepted for Japan, France and Canada. The hypothesis that rates were fully linked cannot be rejected for Italy, the United Kingdom, the Netherlands and Switzerland. The same equations estimated for the 1980s see the γ coefficient fall in value in all instances. It falls to zero for the Netherlands and Switzerland, to about a third of its 1970s' values for the United Kingdom and Canada, and becomes significantly negative for Italy. The coefficient falls slightly for Japan and France, and remains effectively zero for Germany. Estimating this same relationship from January 1986 to the end of the sample period at February 1990 indicates zero or perverse relationships for France, Italy and the Netherlands. In most other cases the value of γ is significantly less than unity, the one exception being Switzerland.

The existence of the EMS and closer monetary policy co-operation between member countries might suggest closer linkages between European rates than with the United States. Tests carried out using Germany as the base country are reported in Table 5. The division of sample periods is now August 1974 to March 1979 (the commencement of the EMS, first period). April 1979 to February 1990 (second period) and January 1984 to February 1990 (the third period in which EMS realignments have been relatively unimportant). In conformity with those in Table 4, the weakest results are again for the second period. The strongest links are between Germany and the United Kingdom, Switzerland and the Netherlands. These results make sense since the exchange rate mechanism of the EMS during the second period, and particularly since the start of the third period, was employed to ensure monetary policies were directed at reducing inflation differentials. France and Italy, as high inflation countries, have been forced to pursue much higher real interest rates at the short end.

If the degree of co-movement between real rates across countries is a measure of the degree of international financial market integration, then the evidence that emerges from these results would not indicate consistent and substantial progress (see Caramazza et al. (1986) for a similar conclusion). Cumby and Mishkin (1986), whose test procedure is employed here, reject the extreme hypothesis of no relationship between real rates in different countries, and also that of fully linked rates across countries, in favour of the conclusion that, for most countries in the sample, the foreign/domestic real interest rate coefficient varies between 0.5 and 0.8. Thus, while there is substantial contemporaneous dependence in short-run real interest rate movements across countries, there remains considerable scope for independent national stabilisation policies and divergence in real rates resulting from asymmetric real shocks.

Longer-term real interest rates require some arbitrary assumption about the treatment of inflation expectations. In Section 2 a three-year centred moving average of inflation was employed. Since unit root tests are relatively robust to short-run measurement error, the results presented in Table 1 are appropriate for considering whether there is a tendency towards long-term real interest parity. These tests suggest in all cases that long-term real differentials possess a unit root – there is no evidence of mean-reverting behaviour.

The failure of real interest parity has been attributed by some (Dornbusch (1976) and Mussa (1983), for example) to sticky prices causing deviations from PPP. Others (Roll (1979); Frenkel (1981); Adler and Lehman (1983); Darby (1983); Mishkin (1984)) infer that deviations from PPP are never reversed or that, equivalently, the real exchange rate follows a random walk. If this is correct, then real interest disparities are permanent, caused by permanent relative goods price movements. Obstfeld (1983), for example, presents an intertemporal maximisation model in which real interest rate disparities are generated by changes in the terms of trade.

(c) The Correlation of Domestic Saving and Investment Rates

This final definition of the degree of international capital mobility was initially proposed by Feldstein and Horioka (1980). A high correlation between national saving and investment implies that the domestic economy cannot tap the world savings pool to increase its level of investment beyond that made possible by the supply of savings from domestic sources. Feldstein and Horioka inferred from their results that a sustained one percentage point increase in the saving rate resulted approximately in a one percentage point increase in the investment rate, which is consistent with the proposition that foreign savings are not internationally mobile. In a recent update of this work, however, Feldstein and Bacchetta (1989) report a savings retention coefficient of 0.79 for the 1980–86 period, which is lower than the 0.91 and 0.86 estimates for the 1960s and 1970s respectively.

Although the Feldstein and Bacchetta paper takes on board most of the criticisms, both theoretical and statistical, that have been directed against the original Feldstein-Horioka paper, domestic saving and investment correlations are still important, though reduced[13]. The overriding issue, however, is whether these results can be interpreted as reflecting imperfect capital mobility. Some economists argue that they cannot (see, for example, Frankel (1989) and Obstfeld (1986)). For the Feldstein-Horioka definition to be a valid measure of the degree of capital mobility, certain necessary conditions are required. First, real interest parity must hold; second, the foreign real interest rate must be determined exogenously to the country in question; and third, all variables that condition the country's investment rate, other than the real interest rate, must be independent of that country's savings rate. When the appropriate instrumental variable technique is employed to deal with this last potential source of bias, the Feldstein-Horioka conclusions remain largely intact. The non-exogeneity of the real rate of interest can be dealt with by using international cross-section data in which the real rate of interest is a constant, and therefore not responsible for the observed savings-investment correlation[14]. This leaves real interest parity. As we have already seen, the bulk of the evidence is unfavourable to the real interest parity hypothesis.

Several models which focus on the distinction between traded and non-traded goods have recently been proposed (see Murphy (1986), Engel and Kletzer (1989) and Wong (1990)) in which it is demonstrated that it is possible to generate a positive correlation between national saving and investment even with fully integrated international capital markets. These results are again unfavourable to the interpretation of the saving-investment relationship as reflecting exclusively the degree of international capital mobility. They serve to demonstrate that the crucial implicit assumption in the Feldstein-Horioka model is that all goods are traded and that PPP for traded goods is fully established within the duration of the typical business cycle. The key assumption is therefore an implicit one about commodity markets rather than financial markets.

Increased capital mobility may, for reasons put forward by Feldstein and Horioka, nevertheless, see some decline in savings and investment correlations. Evidence supporting this can be demonstrated by regressions involving pooled savings and investment data across countries, as in Dean et al (1990). This work is updated and a noticeable decline in the savings/investment correlation can be seen from the evidence presented in Chart 4. While these correlations are an imperfect measure of capital mobility, recent data are not wholly inconsistent with the evidence favouring international financial market integration based on closed and covered interest parity presented earlier.

(d) Overall Assessment

Tests that do not confuse goods and financial market integration and are independent of the accuracy of exchange rate expectations (i.e. tests 1 and 2 in Table 3) show a rapid move to almost complete globalisation of financial markets during the 1980s. While it is an imperfect measure of financial integration, test 5 suggests an increasing trend towards greater independence of national saving and investment in OECD countries. This is consistent with the earlier evidence that cumulated current account imbalances as a share of GDP show no evidence of mean reversion when compared between countries. These findings are entirely consistent with the emphasis placed on the importance of the globalisation process in Section 2.

Finally, if real interest parity were observed empirically, the long-run model in Section 1 would collapse to dependence on developments in cumulated current account imbalances alone. The results in Tables 1, 4 and 5 suggest there is no implication of increasing goods market integration implied by the globalisation of financial markets, so that a potentially important role emerges for real interest differentials in long run real exchange rate determination.


This process began in the mid-1970s with the removal of capital controls in Germany, the United States and Canada amongst the major OECD countries. Liberalisation measures in Japan and the United Kingdom followed at the end of the decade, and France, Italy and some other EC countries have moved steadily towards the complete elimination of controls by the middle of 1990's. See Blundell-Wignall and Browne (1991) for a more complete discussion. [6]

According to Aliber's (1973) definition, political risk has nothing to do with existing capital controls per se, but rather relates to the uncertainty about the intensification or relaxation of future capital controls. For the purposes of the present exercise the distinction between international interest rate differentials arising from these two separate effects is not considered particularly important. [7]

Uninhibited capital mobility is only a necessary condition for closed interest parity. It is not sufficient since the assets in question may not be perceived, for other reasons, as perfect substitutes by market participants. [8]

The duration of the time periods displayed in the graphs coincides broadly with the periods of which the relevant Euromarket has been in existence. [9]

Covered interest parity has some practical advantages over closed interest parity for the purposes of the present exercise. Closed interest parity can only be examined for the limited number of countries for which Eurodeposits are issued in their own currency. Furthermore, for some of these countries the Eurodeposit market is a relatively recent development. These data problems are not as severe for covered interest parity tests. [10]

Covered interest parity in the Eurocurrency market can confidently be regarded as valid. Thus fdm = imt−i*mt where fdm is the forward discount on the domestic currency to maturity m and imt and i*mt are interest rates on domestic and foreign assets with m periods to maturity. Uncovered interest parity says that imt – i*mt = E[Δ(Smt)] where the last expression is the expected change in the exchange rate between t and t+m given information available at t. Assuming covered interest parity to be true, testing for uncovered interest parity is essentially a test for: fdm = E[Δ(Smt)] or equivalently, Fmt = E(St+m) i.e. the m-period forward rate at time t is an unbiased predictor of the future spot rate at t+m. [11]

The absence of a freely-fluctuating market rate for treasury bills in Japan meant that estimates had to be confined to the post-1978 period. Note also that Treasury bills were not issued on a regular basis in Italy before February 1979. The sample division for Japan and Italy is therefore different than that for the other countries investigated. It is respectively May 1978 to May 1984 and November 1979 to January 1985. [12]

Other authors have also reported declining saving retention effects in more recent years. See, for example, Turner (1986), Frankel (1989) and Dean et al (1990). [13]

Even using time-series analysis this issue can be successfully addressed, but is found not to be responsible for the high correlations reported (see Frankel (1989)). [14]