6. Resultados de la prueba piloto 1 Tiempo de trabajo
6.4. Constitución de mecanismos de supervisión sindical
Combination models have been rarely applied to foreign exchange (FOREX) studies and so will form a major part of results presented here. Forecast combination is often used to improve forecast accuracy (Costantini and Pappalardo, 2011). Combining has a long history that predates its use in financial discipline. De Gooijer and Hyndman (2006, 459) stated that “combining forecasts, mixing, or pooling quantitative forecasts obtained from very different time series methods and different sources of information has been studied for the past three decades”. The early contributions of combination methods were made by Bates and Granger (1969), Newbold and Granger (1974) and Winkler and Makridakis (1983). Combining is expected to be useful when the researcher is uncertain as to which forecasting method is best. This may be because the researcher encounters a new situation, has a heterogeneous set of time series or expects the future to be especially turbulent. Clemen (1989) summarised the compelling evidence for the relative efficiency of combined forecasts, in a comprehensive bibliographic review. Combining is most useful when there is uncertainty as to the selection of the most accurate forecasting methods, uncertainty associated with the forecasting situation and a high cost for large forecast errors. Clemen (1989, 559) also reported that “forecast accuracy can be substantially improved through the combination of multiple individual forecasts”. Since then, this same conclusion has been drawn in many academic papers (e.g. Timmermann, 2006; Marcellion, 2004 and Zou and Yang, 2004).
The application of combining forecasts is well researched in the field of geography (Krishnamurti, 1999), media (Langlois and Roggman 1990; Galton, 1878), science (Levins, 1966; Winkler and Pesos, 1993; Weiberg, 1986), engineering (Armstrong et al., 2000), management (Huffcutt and Woehr, 1999), industrial economics (National Industrial Conference Board, 1963; Wolfe, 1966; PoKempner and Bailey, 1970) and tourism (Shen et
al., 2011; Coshall and Charlesworth, 2010; Coshall 2009; Song and Li, 2008). Jore et al.
(2009) examined the effectiveness of recursive-weight forecast combinations for a forecasting output growth, inflation and interest rates. Bjørnland et al. (2010) applied combination models to forecast inflation. Kapetanios et al. (2007) used combination techniques to forecast inflation and output growth. Hibon and Evgeniou (2005) proposed that the accuracy of the selected combinations is significantly better and less variable than that of the selected individual forecasts. Other studies including, Zou and Yang (2004), Terui and Dijk (2002), Goodwin (2000) and Fong-Lin (1998) also confirmed that combination model generates better results than the forecast made by single model.
In Finance, the combination concept also plays an important role. Andrawis et al. (2011) applied combination techniques to forecast NN5 Competition series - a set of 111 time series representing daily cash withdrawal amounts at ATM machines. They combined neural network, Gaussian process regression and linear models via simple average and concluded that combination models improved forecasting performance. Applying the combination approach to data from the NN3 and M1 Competition series, Theodosiou (2011) suggested that a simple combination of four statistical methods produced consistently better results in one-step ahead of monthly and quarterly data. Becker and Clements (2008) examined combination methods to forecasts the volatility of the S&P 500. They found that forecasts based on combination models were the dominant approach. Leung et al. (2001) used investment portfolio returns to combine forecasts. Batchelor and Dua (1995) examined forecasts of real GNP, inflation, corporate profits and unemployment for forecast horizons of 6, 12, and 18 months ahead. Using combination forecasts, they concluded that the mean square error (MSE) of the residuals was reduced by 16.4%. Lobo and Nair (1990) studied quarterly earnings forecasts for 96 firms from 1976 to 1983. Their results showed that combining methods reduced the mean absolute percentage error (MAPE) of the residuals by 5.2%.
Over the past half century, practicing forecasters have advised firms to use combination methods. Fang and Xu (2003) investigated the predictability of assets returns by developing an approach that combines technical analysis and conventional time series forecasts. They concluded that the combined strategies outperform both technical trading rules and time series forecasts on daily the Dow Jones average over the first 100 years. Terregrossa (1999) found that combining financial analysts’ consensus forecasts with a capital assets pricing model (CAPM) stimulates ex-ante forecast leads to superior forecasts of 5 years earning per share (EPS) growth relatively to either component.
There have been very few applications of combining forecasts models in the foreign exchange field. MacDoland and Marsh (1994) applied combination methods in analyses of dollar/sterling, deutschemark/dollar and yen/dollar exchange rate series. They have demonstrated that forecasts made by individual models are not very accurate, are biased and do not take full account of available information. The combining forecasts, however, increased the accuracy of the predictions, but the gains mainly reflect the removal of systematic and unstable bias. Hu and Tsoukalas (1999) examined the out-of-sample forecasting performances of a number of conditional volatility models for a set of 11 European currencies against the German mark. They combined four individual volatility models and concluded that a volatility model outperformed the combination models. Zhang (2003) concluded that for short term (1 month) forecasting of British pound/U.S. dollar, both a neural network model and hybrid (ARIMA and ANN - artificial neural network) models possessed higher forecasting accuracy than the random walk model. For a longer time horizon (12 months), ANN models gave a comparable performance to the ARIMA model. However, the hybrid or combined model outperforms both the ARIMA and ANN models consistently over 1 month, 6 months and 12 months.
Dunis and Chen (2005) investigated the predictive powers of 16 alternative models applied to Euro/U.S. dollar and U.S. dollar/Japanese yen. No single model emerged as an overall optimum of their study. However, ‘mixed’ models incorporating market data of currency volatility, NNR (neural network regression) models and combinations of models performed best in most of the cases. Ince and Trafalis (2006) studied daily values of exchange rates for Euro/U.S. dollar, British pound/U.S. dollar, Japanese yen/U.S. dollar and Australian dollar/U.S. dollar. Using both parametric and nonparametric techniques, they concluded that most of their single or combined models were at least as good as a
random walk forecasting models. Corte et al. (2007) provided a comprehensive evaluation of the short-horizon predictive ability of economic fundamentals and forward premia on monthly exchange rate returns in a framework that allows for volatility timing. They implemented Bayesian methods for estimation and ranking of a set of empirical exchange rate models and construct combined forecasts based on model averaging.
Lam et al. (2008) studied the Euro, British pound and Japanese yen against U.S. dollar. Their empirical results suggested that combined forecasts outperformed the benchmarks and generally yielded better results than simply relying on a single model. Anastasakis and Mort (2009) studied exchange rate forecasting using combination techniques. They applied parametric and nonparametric modeling methods for the daily prediction of the exchange rate market. They concluded that the combined method produces promising results and outperforms individual methods in the case of tested with two exchange rates- the U.S. dollar and the Deutche mark against the British pound. Altavilla and Grauwe (2010) used combination techniques in respect of quarterly exchange rates of the Euro, British pound and Japanese yen against the U.S dollar. They concluded that combining different forecasting procedures generally produced more accurate forecasts than can be attained from a single model. Maté (2011) applied multivariate analyses (especially principal components and factor analysis) to combining forecasts by using daily Euro/dollar exchange rates. Shahriari (2011) and Nouri et al. (2011) conducted a similar study on monthly Iranian rial/British pound exchange rate. Their findings suggest that combination of simple Naїve and cubic regression models fits the data better than individual models.
Combining forecasts is an appealing approach. Instead of choosing the single best model, it is sensible to ask whether a combination of models would help to improve forecast accuracy, assuming that each model has something to contribute. Combining forecasts improves accuracy and there are several ways of combining forecasts. One is to use different data sets and the other is to use different forecasting methods. According to Armstrong (2001, 2), “the more the data and methods differ, the greater is the expected improvement in forecast accuracy over the average of the individual forecasts”. It has been observed in the Finance literature that no single model performs consistently well across all time series and forecasting horizons. Thus, by combining forecasts, the researcher can reduce misspecification bias in the models and increase the prediction accuracy (Theodosiou, 2011). By combining, practitioners can avoid the possibility of choosing the
worst forecasting methods for that particular point in time and hence, robustness the estimations across all forecasting horizons (Armstrong et al. 1983; De Gooijer and Hyndman, 2006). In this present study, the different methods are used to improve forecast accuracy by using combining forecasts. Although statistically based forecast combination methods have had minimal application in the field of exchange rate modelling, the evidence that exists suggests that combination models perform better than the worst single model predictions and sometimes out-perform the best single model (Anastasakis and Mort, 2009).
There have been very few applications of combination models in the foreign exchange field, yet these models have the potential to assist policy makers in making more effective decisions. The use of appropriate combination techniques in exchange rate forecasting is crucial not only for academic researchers but also for practitioners such as governments, banks, insurance companies, businessman, investors, international organisations (IMF, World Bank etc.), tourism authorities, individuals and other related parties such as speculators, hedgers and arbitrageurs. This present study addresses two outstanding issues raised by Poon and Granger (2003). Poon and Granger (2003) highlighted the fact that little attention has been paid to the performance of combination forecasts, since different forecasting approaches capture different volatility dynamics. They also point out that little has been done to consider whether forecasting approaches are significantly different in terms of performance. This study applies the combination forecasting techniques to exchange rate data to fill this major gap of the literature. Although many researchers observe that exchange rates are an important indicator of the economic welfare of any country, most of the studies on forecasting exchange rates are mainly focused on developed and to some extent secondary emerging markets (Abdalla, 2012; Kamal et al., 2012; Hall et al., 2010; Molana and Osei-Assibey, 2010 and Osinska, 2010). However, studies involving emerging and frontier markets are almost non-existent. Therefore, a prime focus of this study is on combination forecasts of each of advanced, emerging and frontier markets’ currencies against U.S. dollar to fill this gap of the existing literature. Furthermore, the majority of studies have concentrated on bilateral exchange rates between advanced countries rather than exchange rates of emerging versus advanced countries and frontier versus advanced countries. This study contributes to the existing literature by assessing the utility of combination techniques in these different contexts.