匯率預測：結合時間序列模型與自注意力機制

Forex Prediction: Combining Time Series Model and Self-Attention Mechanism

簡崇益(Chung-I Chien)

指導教授 ： 林建甫

國立臺灣大學/社會科學院/經濟學研究所/碩士(2022年)

https://doi.org/10.6342/NTU202200714

摘要

匯率預測一直以來都是學界、業界和政府單位關心的議題，匯率的變動可能會影響實體經濟活動和金融市場。過去許多研究顯示，傳統的經濟模型在匯率的樣本外預測方面無法打敗隨機漫步模型，因為傳統模型大多為線性模型，而匯率序列本質上是動態的和非線性的。近年來有許多深度模型被提出，雖然其主要是用在自然語言處理的研究上，但其在捕捉非線性特性的卓越能力，使得越來越多研究開始嘗試將其應用在匯率預測方面。
本文主要參考 Yilmaz and Arabaci (2021)的模型，將匯率的預測拆分成線性和非線性兩個部分，分別使用整合移動平均自迴歸(Autoregressive Integrated Moving Average, ARIMA)模型和自注意力(Self-Attention, SA)機制分別對美元兌加元及澳幣、英鎊兌美元匯率之報酬率進行預測，將此模型與 Yilmaz and Arabaci (2021)研究中使用長短期記憶(Long Short-term Memory, LSTM)架構的 ARIMA-LSTM 模型及隨機漫步模型進行比較後，結果顯示改用自注意力機制的 ARIMA-SA 模型的預測能力較 ARIMA-LSTM 模型來得差，自注意力機制在匯率預測方面無法得到好於長短期記憶機制的結果。甚至，不同於 Yilmaz and Arabaci (2021)所顯示的結果，ARIMA-LSTM 模型在日匯率報酬方面的預測能力亦劣於隨機漫步模型。

關鍵字

匯率 ； 預測 ； 時間序列 ； 深度學習 ； 混合模型

並列摘要

Exchange rate forecasting has long been an issue of interest to academics, industries, and governments, where changes in exchange rates may affect real economic activity and financial markets. Many studies have shown that traditional economic models cannot beat random walk models for out-of-sample forecasting of exchange rates, because most of the traditional models are linear, while exchange rate series are inherently dynamic and non-linear. In recent years, many deep models have been proposed, and although they are mainly used in natural language processing studies, their excellent ability to capture nonlinear properties has led more and more studies to try to apply them in exchange rate prediction.
This paper mainly refers to the model from Yilmaz and Arabaci (2021), which splits the forecast of exchange rate into linear and non-linear components, using the Autoregressive Integrated Moving Average (ARIMA) model and Self-Attention (SA) mechanism to forecast the return of USD/CAD, AUD/USD, and GBP/USD respectively. Comparing this model with the ARIMA-LSTM model using the Long Short-term Memory (LSTM) framework in Yilmaz and Arabaci (2021) and the random walk model, the results show that the ARIMA-SA model has worse predictive power than the ARIMA-LSTM model. Moreover, unlike the results shown by Yilmaz and Arabaci (2021), the predictive power of the ARIMA-LSTM model in terms of daily exchange rate returns is inferior to that of the random walk model.

並列關鍵字

Forex ； Prediction ； Time Series ； Deep Learning ； Hybrid model

參考文獻

Alexander, Don and Lee R Thomas III (1987), “Monetary/Asset models of exchange rate determination: How well have they performed in the 1980’s?”, International Journal of Forecasting, 3(1), 53–64.

Alvarez-Diaz, Marcos (2008), “Exchange rates forecasting: local or global methods?”, Applied Economics, 40(15), 1969–1984.

Dautel, Alexander Jakob, Wolfgang Karl Härdle, Stefan Lessmann, and Hsin-Vonn Seow (2020), “Forex exchange rate forecasting using deep recurrent neural networks”, Digital Finance, 2(1), 69–96.

Dodevski, Aleksandar, Natasa Koceska, and Saso Koceski (2018), “Forecasting exchange rate between macedonian denar and euro using deep learning”, Journal of Applied Economics and Business, 6(2), 50–61.

Engle, Charles and James D Hamilton (1990), “Long swings in the dollar: Are they in the data and do markets know it”, American Economic Review, 80(4), 689–713.