Long Term Memory in Foreign Exchange Market Returns: International Evidence.

Author:Jain, Anuradha
 
FREE EXCERPT

INTRODUCTION

A financial market is said to be efficient if it discounts all information available in the public domain. The faster the absorption of information in compensating price differential, the more efficient is the market. Contrary to efficient market hypothesis, observations have been made that financial market returns have positive auto-correlation in short term and negative auto-correlation in the long-run (Poterba & Summers, 1988). The meaning of negative autocorrelation can be evaluated as the presence of long-term relationship or memory among variables, as negative correlation suggests a situation of mean-reverting. If two variables are having a long-term relationship, the future price predictions of the variables become relatively easier with low errors. But, this statement goes in a direction opposite to the proposition of the Random Walk Model, which suggests that all possible information available gets factored in the price of the asset and, hence, predicting the future value of the asset is not possible. But, Skjeltorp's (2000) study suggests that when new information arrives in the market, some market participants react immediately, while some wait for some more or relevant information and, hence, their actions get accumulated which creates an eruption in the market. Cheung (1993) also identified long-term memory in the foreign exchange market, while Barkoulas et al. (2003) concluded that exchange rates are best characterized by short-term memory dependence process rather than long-term memory dependence process. Diebold and Rudebush (1989) suggested that economic fundamental variables have long-term memory property and, hence, foreign exchange being one of the most fundamental economic variables is also expected to show long-run memory property. Some of the studies have taken different forms of the returns to evaluate the long-rum memory process. For example, Hiemstra and Jones (1997) identified that returns of the stock index do not show any evidence of long-term memory process, while the variance of the returns shows this long-term memory creation process. Commonly, R/S test is used to evaluate the long-term memory process, but it has a number of limitations: conventional R/S analysis using a Hurst regression can be biased toward accepting a long term dependence hypothesis even when the true process is first order autoregressive (Harte, 1987). Lo (1991) developed a modified R/S analysis to overcome the limitations of prior model. But, there is a limitation with the modified R/S method also, as it assumes an infinite memory process. The modern econometric tool provides a solution to this. Johansen Co-integration test can be used to identify any long-term relationship between variables. For, Johansen Co-integration, the necessary condition is that all variables must be stationary at order I(1). This paper has taken three exchange rate markets into consideration to evaluate the long-term memory process. The three markets under consideration are INR_USD, EUR_USD and GBPUSD exchange rate markets. The three markets taken into consideration are important for study because of higher convertibility and liquidity in foreign exchange markets. These markets are also important, as recently these markets have seen a huge change in economic settings in general and in financial settings in particular. The high level of variance in USD_INR (see Table 1 below) also suggest that during these period 2008-2012, Indian currency saw a high volatility compared to other two currencies taken into consideration.

The study is also important at this point, as after a financial crisis or economic turbulence the entire economic variables change their equilibrium positions and try to re-establish a new equilibrium. Hence, to identify the long-term memory process in foreign exchange market is very important after a recent crisis. The long-term memory process can help multinational companies forecast their obligations with lower error and, hence, make better predictions. This paper has been divided into the following categories: Section II describes the literature review, section III deals with...

To continue reading

FREE SIGN UP