9   Conclusions and future research

9.1 Conclusions

Air travel demand forecasting methods are widely investigated in literature. However, the unsatisfactory forecasting performance still exists and leads to poor effectiveness of the business management decisions. Nowadays, the improved computational capabilities are making the adoption of more sophisticated methods possible, and massive amounts of information are becoming available from various data sources, which should be utilized properly to improve the demand forecasting performance. All these factors make this book necessary and meaningful.

In this book, we provide comprehensive research on air travel demand forecasting methods, for various forecasting time horizons ranging from the short term (less than 1 year) and the medium term (3 to 5 years) to the long term (longer than 5 years). Based on the basic idea of TEI@I methodology, we firstly propose a novel and unified air travel demand forecasting framework, which incorporates both the structured data sources and the unstructured data sources such as the Internet. And experts’ knowledge will guide us in all of the forecasting processes. Forecasting method studies in this book are all based on this unified framework.

In the first part of this book, we firstly introduce the main motivations of this book, describe the three main forecasting problems that we focus on in the rest of this book and outline the whole structure of this book. Then, we review the existing research about demand forecasting in a two-step way. In the first step, we focus on the general demand and classify these previous studies by the forecasting time horizons, i.e., short term, medium term and long term, to review the common methods applied in different forecasting time horizons. In the second step, we focus on the air travel demand and discuss common forecasting methods for various categories of air travel demand, including the air travel demand of O-D pairs, air travel demand of an airport and nationwide air travel demand. In addition, we also summarize the main determinants for the general air travel demand to provide a guide for the studies in the rest of this book. After reviewing the existing research, we introduce the theoretical basis of this book, i.e., TEI@I methodology, which was originally proposed for analyzing complex systems due to the inherent complexity and nonlinearity in a complex system. Along with the introduction of the theoretical framework and the unified air travel demand forecasting framework, we also briefly cover several common forecasting methods and models, including econometrical forecasting models such as the ARIMA/SARIMA models, VAR/VEC models and ARCH/GARCH models, as well as several artificial intelligence techniques such as the ANN, SVR/LSSVR and GP models. We also introduce the main types of combination forecasting methods, and the key issues involved in the judgmental adjustment with expert knowledge. To sum up, in the first part of this book, we introduce the existing research and common forecasting methods in a comprehensive way. It can provide a reference for both researchers and practitioners.

In the second part of this book, we provide several studies about air travel demand forecasting for various purposes and forecasting time horizons according to the forecasting problems raised in the first part. Firstly, we present a scientometric analysis of the general demand forecasting literature (1975–2015), to investigate the whole landscape of the present demand forecasting research and to identify main thematic patterns, landmark articles and emerging trends for future research with a visual description. The scientometric analysis results show that the future research in demand forecasting mainly focuses on the combination forecasting techniques and hybrid intelligent forecasting system. This is the first study which adopts a scientometric analysis method in demand forecasting literature. Secondly, we propose several suitable forecasting methods for different forecasting problems. In Chapter 5, we mainly focus on the model selection and construction of the short-term air travel demand for the complex and volatile economic circumstances, and we propose an integrated short-term forecasting framework based on the TEI@I methodology, with the empirical mode decomposition method as a decomposition method. The final empirical results, based on historical air travel data of the Hong Kong International Airport, confirm the superior forecasting performance of our proposed forecasting method. In Chapter 6, to deal with a poor forecasting performance problem for those months suffering moving holiday effects, we propose a forecasting framework based on the seasonal decomposition method with a novel use of Google Trends data. The Google Trends data are creatively used as an explanatory variable to capture the moving holiday effects. In addition to a forecasting model, we also propose a nowcasting method for the monthly air travel demand with the weekly Google Trends data. The empirical results show that the information of a higher frequency in the current period can effectively help to improve the forecasting accuracy. In Chapter 7, we propose a demand forecasting method based on the stochastic frontier analysis method and a model average technique, to deal with an unconstrained demand forecasting problem. Unconstrained demand forecasting is often difficult due to the unobservability of the applicable historical demand series. Hence, the main objective of this chapter is to develop a method that can estimate and forecast the unconstrained demand, i.e., the ‘true demand’, properly and scientifically. This method solves a long-existing problem that has always been ignored in demand forecasting literature, and considers the demand forecasting problem from a new perspective. Then we also implement an empirical application of this method on the medium-term air travel demand forecasting. The empirical results show that in addition to its ability to estimate unconstrained demand, our method outperforms other common forecasting methods in terms of forecasting passenger traffic. And finally in Chapter 8, we deal with a long-term air travel demand forecasting problem. Forecasting the future air travel demand over a time horizon longer than 5 years usually faces many challenges that have never been faced in the short-term or medium-term forecasting. Hence, to fulfill such a complex job, we propose an integrated method for long-term air travel demand forecasting in China mainly with an ARDL bounds testing approach to cointegration and scenario planning technique. Within this forecasting framework, we can rely on not only China’s own development experience (i.e., the historical data pattern) by exploring the long-run relationship between demand and its drivers, but also learn the development experiences from other developed air travel markets with the logistic curve fitting technique.

9.2 Future research

The future research will be mainly focused on the following four directions.

Firstly, we should seek more sophisticated economic and statistical theories, and better artificial intelligence algorithms, because demand forecasting is far from easy, and the research about demand forecasting methods is far from mature.

Secondly, we should pay more attention to experts’ domain knowledge and forecasting experience. How to reasonably and scientifically incorporate experts’ knowledge in a forecasting process is still a big challenge. There are problems including: i) How to evaluate the value of both historical information and experts’ domain knowledge, i.e., how to decide the proportion of the two aspects in a forecasting process? ii) How to combine the two aspects, to achieve the final forecasts. Here, the ‘experts’ denote not only academic researchers but also the practitioners from industries.

Thirdly, when doing demand forecasting research, academic researchers should value the opinions of practitioners from industries, and find more valuable forecasting problems faced by those practitioners.

Finally, to exploit the availability of more plentiful data observations, we should develop more valuable data sources and a specific data processing method for the ‘big data’. Data sources such as the search engines should have more attention paid to them because they link tightly with the consumers’ daily life and interests.