Based On EEMD-ARIMA, Hotel Occupancy Forecasting Research

Tourism industry has been becoming one of the most developed industries after the World War ?. Known as "smokeless industry", it gives a great impetus to the economy, so every country and regionattaches importance to its development. But tourism industry is fragilemeanwhile; it is usually affected by natural environment, economic environment, humane environment, policies and natural disasters. Preparedness ensures success; tourism related supervision departments and operators should not only enforce supervision and management to tourism industry, but also pay attention to tourism demand forecasting, by which can allocate resources rationally and avoid risk. To forecast tourism demand accurately has been being an important topic in tourism research, but due to the unstableness of tourism market and complexity of casual factors, tourism demand forecasting is difficult relatively. However, sustained and stable development cannot be ensured without the support of precise demand forecasting. As one of the three pillar industries in tourism, hospitality industry has strong influence on tourism industry, and it is also caused by many factors. The accurate forecasting also supports the management and operation of hospitality industry.Choosing tourism demand as perspective, this research uses hotels in Charleston area as empirical cases, and aims to establish a new forecasting model by EEMD and ARIMA for hotel occupancy of Charleston, East Cooper and North Charleston area and give better forecasting results. We choose Autoregressive Integrated Moving Average (ARIMA) model as a basis of forecasting model; first, use R to establish three ARIMA models of three areas relatively, and analyze 52-week forecasting results, these models and results are the basis of further research; then by introducing a new adaptive spectral analysis method-ensemble empirical mode decomposition (EEMD), hotel occupancy signals are filtered by EEMD to several singals that called IMFs to explore the internal fluctuation rule of data. Based on this, we put forward a new method that applies decomposition and combination ideaby combining high-frequency fluctuation, low-frequency fluctuation and trend from IMFs; then model these three signals to ARIMA models. Study results show that by using EEMD-ARIMA model to forecast hotel occupancy, medium term forecasting (26 weeks) reveals better forecasting results than ARIMA model. Specifically, in medium term forecasting, EEMD-ARIMA model reduces 31.25%,14.9% and 1.67% error respectively from ARIMA in MAPE test, and 31.03%,19.7% and 4.1% respectively in RMSE test; in long term forecasting (52 weeks), EEMD-ARIMA model reduces 9.67% error in MAPE and 16.4% in RMSE. In general, EEMD-ARIMA model is more robust than ARIMA model in forecasting hotel occupancy. Through this research, main conclusions are as follow:(1) ARIMA models have well results in both medium term and long term forecasting. Bymodelling hotel occupancy data of these three areaswith ARIMA, we get general time series forecasting models. Model testing tool MAPE and RMSE show that ARIMA models achieve good effect. However, only ARIMA model cannot catch every detail in non-linear or strongly fluctuated time series, so ARIMA model still remains to improve in forecasting.(2) By EEMD, fluctuated details can be extracted accurately from hotel occupancy signals; on one hand, it helps researcher to analyze fluctuated rules in different frequencies; on another, it exposes the trend and developing pattern by t signal. The study shows that the internal developing pattern is in coincidence with economic developing status.(3) EEMD-ARIMA models can enhance forecasting accuracy. Because EEMD can decompose data deeply, so combined with the EEMD method, EEMD-ARIMA model decomposes the complex data-hotel occupancy, and smoothes fluctuation; then combines several fluctuated signals into high-frequency fluctuation, low-frequency fluctuation and trend (t signal), which can help regulating the consistency of data. Eventually, EEMD-ARIMA model shows better forecasting results and more robust index in MAPE and RMSE test; EEMD-ARIMA models enhance accuracy in medium term forecastingin all three empirical cases.Innovations are as follow:(1) Based on ARIMA model, this research push common time series research to deeper area, which studies signal decomposition, broadens research methods and enriches research contents of time series.(2) This study first introduces EEMD to tourism time series research, which was still a blank space in tourism study. On the basis of traditional time series methods, this study adds EEMD method, and proposes a new EEMD-ARIMA forecasting model. Based on existing model theories, the new model takes the particularity of the empirical cases into account, applies rationalism to experience, and comes to a conclusion that new model do can improve forecasting ability of traditional time series.