Forecasting U.S. Tourist arrivals using optimal Singular Spectrum Analysis

Abstract

5This paper introduces Singular Spectrum Analysis (SSA) for tourism demand forecasting 6 via an application into total monthly U.S. Tourist arrivals from 1996-2012. The global 7 tourism industry is today, a key driver of foreign exchange inflows to an economy. Here, we 8 compare the forecasting results from SSA with those from ARIMA, Exponential Smoothing 9 (ETS) and Neural Networks (NN). We find statistically significant evidence proving that 10 the SSA model outperforms the optimal ARIMA, ETS and NN models at forecasting total 11 U.S. Tourist arrivals. The study also finds SSA outperforming ARIMA at forecasting U.S. 12Tourist arrivals by country of origin with statistically significant results. In the process, we 13 find strong evidence to justify the discontinuation of employing ARIMA, ETS and a feed-14 forward NN model with one hidden layer as a forecasting technique for U.S. Tourist arrivals 15 in the future, and introduce SSA as its highly lucrative replacement. rather than weakened this need to forecast tourist demand accurately. 25As discussed in the following section there is an extensive and high profile existing literature 26 on forecasting tourism demand. This literature covers a wide range of different forecasting 27 techniques, applied to a wide range of different countries or locations. The purpose of this paper 28 is to add to this literature by introducing a new model for forecasting tourist arrivals and to 29 apply it to inbound U.S. Tourist arrivals. Forecasting U.S. Tourist arrivals is both a demanding 30 and important task, mainly because these data exhibit a high degree of fluctuation over time. 31Figure 1 depicts the time series for total monthly U.S. Tourist arrivals between January 1996 and 32 November 2012. A first look at the time series suggests signs of seasonality in U.S tourist arrivals. 33The figure also shows that the tourism industry in the U.S. is experiencing rapid development in There are a number of components which define a good demand forecasting model for tourism Australia, concluding that using models expressed in first differences increased forecast accuracy. Using data for Hong Kong they find use of tourism arrivals to be more affected by income in 117 the country of origin and tourism expenditure to be more sensitive to prices. Wan, Wang, and 118Woo (2013), also using tourist arrival data for Hong Kong, assess the properties of disaggregated 119 forecasts using a seasonal ARIMA model relative to aggregate forecasts. They find the sum of 120 disaggregated forecasts to provide greater accuracy than an aggregate forecast. 121A very closely related strand in the literature seeks to combine two or more forecasting 122 models into a new hybrid model and to test whether this results in greater forecast accuracy. 123Andrawis, Atiya, and El-Shishiny (2011) finds that, in forecasts of tourism arrivals into Egypt, 124combining short and long term forecasts improves accuracy compared to the individual forecasts. 125Cang (2011)

New developments in tourism and hotel demand modeling and forecasting

“…For example, Divino and McAleer (2010) Very recently, a nonparametric forecasting technique, the singular spectrum analysis (SSA), has been introduced into the tourism literature (Beneki et al, 2012;Hassani et al, 2015). Assuming that a time series consists of signal and noise, unlike traditional time series models which forecast both signal and noise, SSA aims to filter the noise and forecast the signal only (Hassani et al, 2015). Similar to a STS model, SSA decomposes a time series into independent components such as trend, seasonal and business cycle components but, as a nonparametric method, SSA is model-free and data-driven, making no assumptions about the data-generating processes.…”

Comparative analysis of exponential smoothing models to tourists' arrivals in Serbia

“…The weights decline exponentially over time, since the most recent data are considered to be more influential on forecasts than are older observations (Coshall, Charlesworth, 2011). Exponential smoothing model considers the error, trend and seasonal components in choosing the best model by optimizing initial values and parameters (Hassani et al, 2015).…”

Intracranial Pressure Forecasting in Children Using Dynamic Averaging of Time Series Data

“…In this approach, the most recent observations were weighted higher than the previous ones. Therefore, the ETS model can produce smarter forecasts compared to the average method [18,19]. In order to forecast the next 30 min of a patient ICP, both the ARIMA and ETS models were fitted to the ICP time series data.…”