|
Introduction: the macro economy, international financial scholars are always on the exchange rate there is a lot of interest, because of the long time since the projected performance of the exchange rate based on the the macroeconomics exchange model is not working well. Theoretically determine the exchange rate macroeconomic variables such as monetary aggregates, real income, interest rates, inflation, etc., in the empirical analysis, it is difficult to reflect the decisive role of these variables on the nominal exchange rate, which is a mystery in the field of international finance. Meese and Rogoff in their articles (1983a and 1983b), the first time that such a result: empirical analysis based on samples of rms standard structural exchange rate model in the 1970s as well as a time series model to predict the performance of the exchange rate are poor, far better than a simple random walk model. The results show that the nominal exchange rate tends to be statistically a random walk process, but decided to exchange rate changes in macroeconomic variables, macroeconomic theory, but does not tend to be a random walk process. In the decades after, improve macroeconomic model, beyond the simple random walk model in the exchange rate forecasting performance goals yet to achieve. Recently, however, the exchange rate field with many new discoveries aroused widespread controversy, for example, be used to analyze the structural time series model of the real exchange rate determinants that real shocks explain the real exchange rate fluctuations is a major and significant reasons. These findings suggest that the exchange rate fluctuates around its equilibrium value, it may be caused by factors not considered by some macroeconomic model. This paper attempts to identify such factors, factors such as historical data to test the possibility of improving the exchange rate model for the prediction of the nominal exchange rate. In this paper, we consider the price of oil is one of the factors, oil prices affect macroeconomic variables, such as income, account balance and savings, these variables will affect the stock and distribution of assets between the oil-importing countries and oil-exporting countries, which break the original market equilibrium, caused by changes in the exchange rate. For decades, scholars have studied the relationship between oil prices and the exchange rate, Robert A.Amano and Simon van Norden, in his article (1993) proved that the real exchange rate and the price of oil cointegration and causal relationship between oil prices flows to the real exchange rate. Although many articles have studied the relationship between oil prices and exchange rates, but try to use oil prices to forecast the exchange rate or to improve the exchange rate forecast article, but it is rare. Oil prices is interpreted as a new variable added to the exchange rate model, collecting historical data and empirical analysis to test whether the price of oil can improve these models against the U.S. dollar / mark, dollar / pound, dollar / yen three types of rates forecast. In this paper, we consider the model two macroeconomic models: Flexible Rate (Frankel - Pilsen) model and sticky price model (Dornbusch - Frankel), as well as a simple random walk model. Model: the first model is the flexible price model proposed by Frankel and Plzen, the model can use the following equation to express: s t is the logarithm of the exchange rate, the m t and m t * denote the logarithm of the domestic money supply and foreign money supply, y t and y t * denote the logarithm of the national real income and foreign real income the r t and r t * representing domestic and foreign short-term interest rates, μ is the error term. deviation from purchasing power parity equation, the model extends the the Frankel - Pilsen model, adding a new variable: domestic and foreign long-term inflation rate differential. Where π e andπ e * , said the difference between the domestic and foreign expected inflation rate. In addition to the two macroeconomic models, we also consider a simple random walk model, as shown in the following equation: which the future rate value S t exchange current value the s t < / sub> plus the error ε t 1 constitutes the error term is a White noise, the standard normal distribution, expectation 0 and variance 1). Article reason to consider a simple random walk model, because the model does not contain any other explanatory variables, changes in the exchange rate is only determined by a random factor, but there are a number of explanatory variables macroeconomic model, if the price of oil in improving macroeconomic model forecast ability performance is not satisfactory, we can also refer to the price of oil in improving the performance of the random walk model predictive ability, because the former may be caused due to the price of oil and other explanatory variables, multiple linear, the latter is able to respond to oil prices the decisive role played by the exchange rate change as the only explanatory variable. Decide in what form the oil price interpreted as a new variable is added to these models, we consider three alternative forms: the logarithm of the price of oil, the price of oil, as well as the growth rate of oil prices. We found that the growth rate of the oil prices and exchange rates have a good linear relationship, but also small and other explanatory variables, the growth rate of the price of oil so the article selected as a new variable, the new variable coefficient can be interpreted if the oil price increase of 1%, the exchange rate will change how much per cent. Added a new variable flexible price macroeconomic model can be expressed by the following equation, adding new variables sticky price model can be expressed by the following equation: where P represents the new variables, the percentage increase in oil prices. This simple random walk model to add a new variable corresponding model can be expressed by the following equation: where △ p t 1 is the percentage increase of the oil price, ε t 1 is the error term of the model is in fact a impermanence several simple linear regression model. The exchange rates in all models appear in logarithmic form, so that the units can be obtained statistical variables, making comparability between the different models. Methods: All models will be used to dollar / mark, dollar / pound sterling, the U.S. dollar / Japanese yen exchange rate forecast, prediction intervals of one month, six months and 12 months. The data used in this paper is the month observational data, For each exchange rates, the use of data samples have during the corresponding year span. Macroeconomic model, the only macro variable data exist to its regression prediction, due to limited resources, this article can be collected to the limited data available, the exchange rate for the U.S. dollar / mark, the the regression data sample interval from January 1986 to In April 1997, for the dollar / pound exchange rate, the regression data sample period January 1987 to December 2008, for the dollar / yen exchange rate the regression data sample interval for the January 1986 to April 2009. This paper is the rolling regression method, i.e. for the same linear regression equation, using the rolling point of a constant number of data, multiple regression, to obtain multiple coefficients, the multiple prediction methods. For example, for the dollar / pound exchange rate in case prediction interval of one month, with the first 60 months (from January 1987 to 1991), the data for the first time regression to generate variable regression coefficient of the regression model was used to predict the rate value of the next stage, i.e. into the January 1992, the value of the individual explanatory variables can be; the next stage of a regression, in January 1992, the data is added to the sample, while the 1987 January data were excluded from the sample at this time the number of observation points at 60, these data again for the same equation of regression, a new variable regression coefficients, and then into the next stage of the observation points numerical February 1992 data, the predictive value of the model for the exchange rate in February 1992; and so forth until the rolling regression has been forward to the last observation points to the total sample. Through a rolling process of return, we can get a series of regression coefficients of the dollar / pound exchange rate, corresponding to 204 the next phase of the exchange rate from the predicted values. Not only to test the performance of the models in a prediction interval, also tested the predictive power of prediction interval under each model in six months and 12 months. Different prediction intervals because test whether oil prices can improve the predictive ability of these models in the longer forecast period. Testing forecast accuracy generally two standard a sample accuracy standards, another sample accuracy standard, the former measured data samples for regression fit of the regression model, such as the R square , while the latter is measured in the regression model predicted performance data samples for regression. In this paper, the accuracy of forecasting using samples to evaluate the predictive ability of each model and adding a new model of oil price growth rate variable exchange rate. Meese and Rogoff in their article, the evaluation sample forecasting ability of the three parameters: the root mean square error to compare the model to predict the performance of the main parameters, the average error and the average absolute error for the auxiliary parameters, this article only gives the root mean square error consistent results, because the results shown by the results and root mean square error of the average error and the mean absolute error. Where k = 1, 3, 6, 12, on behalf of the forecast interval, that is, one month, six months and 12 months. h represents the width of the rolling regression window, that is, the number of observation points for each regression, N k is the total number of regression, F i (hi × j) with the first i regression equation for point-in-time (hi × j) exchange rate forecast value. Every regression is used to calculate the predicted values ??of the next to the k-th point in time exchange rates, the last return is not used to generate the exchange rate prediction value. N k * k representatives to generate the total number of the exchange rate forecast values. In addition to test the predictive ability of these models on the exchange rate in the total sample interval, we also tested the model in some sub-sample interval predictive ability of the exchange rate of the three models, specifically, the paper examines the two types sub-sample, one subsample showing a rising or declining trend in oil prices, another common volatility of oil prices and exchange rate trends are significant sub-sample case. The first sub-sample period from January 2003 to September 2008, the continued rapid rise in this period, the price of oil, the price of crude oil rose from $ 33 a barrel to $ 129. Second sub-sample intervals different exchange rates vary, depending on the time span of the data collected, the macroeconomic model, the dollar / mark exchange rate subsample interval from 1986 to 1997, the U.S. dollar / Euro exchange rates sub-sample interval From 2001 to 2008, the dollar / yen exchange rate sub-sample period from 1997 to 2008; random walk model, the dollar / mark exchange rate of the second sub-sample interval from 1986 to 1999 for the dollar / pound exchange rate From 2001 to 2009, the U.S. dollar / yen exchange rate from 1997 to 2009. Results: For each model, each exchange, and each of the prediction interval, adding oil prices RMS error reduction in the column in the following table. Two macroeconomic model, the flexible price model and the sticky-price model, the price of oil failed to improve the predictive ability of the two models of the exchange rate. For simple random walk model, the new model in oil prices can really make the root mean square error decreased, but the magnitude of this reduction is very small reduction in RMSE of less than 0.02, the forecast increase in accuracy of less than two percent reason: two macroeconomic models, flexible pricing models and sticky price model, the addition of oil price growth rate variable, the predictive ability of the exchange rate is still not improved, this article is to proceed from the following aspects, analysis of the results. Error caused by the sampling error is the only observed part of the sample rather than the overall sample, the experiment, the sample size of each exchange were: the dollar / mark exchange rate of 136 observation points 264 observation points, the dollar / pound exchange rate, the dollar / 192 yen observation point, these samples is very small, the sampling error may be significant. Simultaneous equation bias also contributed to the results of one of the possible reasons for this deviation is the error term u and the independent variable exchange rate may cause errors. This article was used to predict the data is the history of the true value for each explanatory variable, rather than at the time of its expected value, and this is one possible explanation of the cause of error. Another possible reason is related to oil prices and other explanatory variables multicollinearity relationship, in this experiment, the price of oil and other explanatory variables may be highly correlated, which contains information other explanatory variables, in turn, other explanations the variable contains oil prices in oil prices this variable to the macroeconomic model, the root mean square error does not significantly improve. However, multicollinearity test results do not support this assumption, according to function of splus collinearTest, oil prices and other explanatory variables collinearity test results: for the dollar / yen exchange rate, test results 15.72818 for dollars / mark exchange rate, the test results for 4.993924, the U.S. dollar / sterling exchange rate, the test results of 5.380801. Theoretically collinearity test result greater than 20 was sufficient to show significant multicollinearity exists. For simple random walk model forecast interval of 12 months, the price of oil and failed to improve its predictive power of these three types of rates, but if you take into account the fact that, should oil prices, adjusted for changes in exchange rate process is fast and adjustment mode is constantly changing, so the result is perhaps not surprising, because the previous 12 months the exchange rate historical data to predict the current exchange rates, the predicted results unsatisfactory imagined. Conclusion: For each exchange, in each of the forecast period, the two oil price variable macroeconomic model, flexible price model and the sticky-price model, have failed to increase in the exchange rate forecasting accuracy. The use of the model of the sub-sample data also failed so that the RMS error is significantly reduced. But whether this result can be attributed to sampling error, multicollinearity force equation of deviation, the price of oil and other explanatory variables, we still can not be asserted. Oil prices variable random walk model does increase in the exchange rate forecasting accuracy, but this increase is limited to one month, six month prediction interval and insignificant, the root mean square error reduced to 0.02, ie joined the oil price forecast in the six month intervals, the random walk model than the original model, the root mean square error in the forecast on the dollar / pound sterling exchange rate 0.02. Prediction interval of 12 months, oil prices have failed to improve the predictive ability of the random walk model. Taking into account the rapid changes in exchange rates, this projected performance is not surprising, because it is the use of data to make predictions in the previous 12 months. According to the experimental results, the overall price of oil in improving the role played by the predictive ability of the existing model of exchange rate is limited. The inadequacies of this article include: no explanatory variables using the expected value, the true expected value of the variable to be predicted using seasonally adjusted data, use relatively small samples, and used to compare the model to predict the ability standard suitable and so on. We will be the focus of future research on the following: the recent sharp decline in oil prices the relationship between oil prices and exchange rates, oil prices and the exchange rate of the non-linear relationship, using the outside of the root mean square error predictive capability evaluation criteria, such as exchange rate the accuracy of the prediction of the change in direction.
|