Untitled

Abstract

Abstract:In this paper, we present the result of our experiment on analyzing a time series data such as the Nikkei 225 index or foreign currency exchange USD/JPY past data and text corpus related to Japanese economy and finance news or reports. There are several researches reported that forecasting of time series with additional features based on text data can be beneficial rather than relying on time series data history only. Experts on investment are usually making their decision based on those text data as they found patterns on them called the fundamental analysis. Time series analysis based on past history only are called technical analysis. The combination of them should make a better prediction system. The text data also provides an explanation or indication to the trend patterns (uptrend, downtrend, or trend reversal), important moves (spikes, releases) and/or volatility. We collected all Nikkei 225 index and USD/JPY past data on daily average closing prices, all Bank of Japan monthly reports and all related available news text data. We applied on the time series Nikkei 225 data the technical analysis such as simple moving average (SMA), regression, ARIMA models and seasonal return rates, GARCH model combined with text data features such as topic models. We aggregate topics yearly or a long period of 5 to 10 years. Topics are extracted with topic modeling techniques such as Latent Dirichlet Allocation (LDA) or Non-negative Matrix Factorization (NMF). Through the observations of those technical and fundamental analyses, we show our results and conclude with some forecasts data.