Lin, Goodwin and Song (2014) showed that Delphi-based expert adjustments of econometric forecasts of visitor arrivals to Hong Kong improved accuracy on average, yet were biased for individual origin markets and degraded accuracy for the USA and helped least for Australia, where the statistical forecasts were already highly accurate (MAPE below 3%), and they raised three linked questions for future research: under what conditions and for which market types judgmental adjustment should be withheld; whether mechanical integration of independent statistical and judgmental forecasts outperforms the voluntary (Delphi-based) integration used in the Hong Kong Tourism Demand Forecasting System; and whether structured debiasing mechanisms (feedback on past errors, explicit quantification of event impacts, alerts to systematic tendencies) would reduce expert bias in such systems. The subsequent literature provides only partial progress — the same first author linked adjustment benefit to series variability on the same single-destination data, hotel revenue-management research documented context-dependent override value, and combination frameworks now integrate expert judgment into tourism forecasts — but no controlled comparison of mechanical versus voluntary integration in tourism demand forecasting, no test of debiasing mechanisms in a tourism forecasting support system, and no systematic characterization of withholding conditions beyond one destination, one expert panel, and one five-quarter period. The problem — establishing when expert judgment should enter statistical tourism demand forecasts, and in what form — remains open.

Background

Tourism demand forecasting predicts visitor arrivals (or expenditure) from origin markets to a destination, most often with econometric time-series models; the study at issue used an autoregressive distributed lag error-correction model (ARDL-ECM), a standard specification that relates demand to origin income and own/substitute prices while separating short-run dynamics from long-run equilibrium. In operational practice, statistical forecasts are rarely used raw: experts revise them to incorporate non-time-series information such as planned events, policy changes, or crises. This is called judgmental adjustment. The Hong Kong Tourism Demand Forecasting System (HKTDFS) is a web-based forecasting support system in which a panel of academic and industry experts adjusts econometric baseline forecasts through the Delphi method — a structured, anonymous group-elicitation procedure in which panelists submit independent estimates over iterative rounds and receive aggregated feedback between rounds. Goodwin (2000) distinguished two ways of bringing judgment and statistics together: voluntary integration, where experts are supplied with statistical forecasts they may ignore, accept, or adjust (the HKTDFS design), and mechanical integration, where independent statistical and judgmental forecasts are formally combined, for example by averaging or Bayesian methods. Accuracy is conventionally measured by the mean absolute percentage error (MAPE) and related scale-free error measures, and bias is tested by regressing realized outcomes on forecasts. Prior work by the same team (Song, Gao and Lin 2013; Lin 2013) found that Delphi-based voluntary integration improved Hong Kong tourism forecast accuracy, particularly at short horizons; the Delphi elicitation tradition in tourism forecasting is surveyed in Lin and Song (2015). Lin, Goodwin and Song (2014) then evaluated accuracy and bias of the adjusted forecasts themselves: on average the adjustments helped, but they were biased at the individual-market level (overestimation for Australia, Taiwan and the UK; underestimation for fast-growing Mainland China, consistent with underadjustment of steep trends), and they harmed accuracy for the USA and helped least for Australia, where the statistical forecasts already had MAPE below 3%. The unresolved question this raises — when expert intervention should be withheld, whether mechanical integration would do better, and whether built-in debiasing would curb the market-specific biases — has not been answered by the subsequent literature described below.

Problem Statement

The problem, as raised by Lin, Goodwin and Song (2014) and left open by the subsequent literature, concerns determining when and how expert judgment should enter statistical forecasts of tourism demand, and comprises three linked questions: (1) Under what conditions and for which market types should judgmental adjustment of statistical tourism demand forecasts be withheld? The conditions of interest are those the source evidence implicates and field conventions make operational: maturity and haul type of the origin market, the initial statistical forecast's own accuracy (e.g., MAPE below 3% already at baseline), series variability, trend steepness, exposure to special events, and forecast horizon — so the question is where adjustment improves accuracy, where it is neutral, and where it degrades it. (2) Within that boundary, does mechanical integration — formally combining independent statistical and judgmental forecasts by averaging or Bayesian methods — outperform the voluntary, Delphi-based integration used in the Hong Kong Tourism Demand Forecasting System? (3) Do structured debiasing mechanisms built into such forecasting support systems — provision of feedback on past errors, requiring experts to explicitly quantify the impact of forthcoming events, and alerting forecasters to their systematic tendencies — reduce the market-specific biases (over- and under-forecasting tied to trend characteristics) that the source study documented? The three questions jointly address when expert intervention should be withheld, and, when it is admitted, in what form (integration mode) and under what safeguards (debiasing design).

References

  1. Lin, V. S., Goodwin, P., & Song, H. (2014). Accuracy and bias of experts' adjusted forecasts. Annals of Tourism Research, 48, 156–174. DOI: 10.1016/j.annals.2014.06.005. https://doi.org/10.1016/j.annals.2014.06.005
  2. Goodwin, P. (2000). Correct or combine? Mechanically integrating judgmental forecasts with statistical methods. International Journal of Forecasting, 16(2), 261–275. DOI: 10.1016/S0169-2070(00)00038-8. https://doi.org/10.1016/S0169-2070(00)00038-8
  3. Song, H., Gao, B. Z., & Lin, V. S. (2013). Combining statistical and judgmental forecasts via a web-based tourism demand forecasting system. International Journal of Forecasting, 29(2), 295–310. DOI: 10.1016/j.ijforecast.2011.12.003. https://doi.org/10.1016/j.ijforecast.2011.12.003
  4. Lin, V. S. (2013). Improving forecasting accuracy by combining statistical and judgmental forecasts in tourism. Journal of China Tourism Research, 9(3), 325–352. DOI: 10.1080/19388160.2013.812901. https://doi.org/10.1080/19388160.2013.812901
  5. Lin, V. S. (2019). Judgmental adjustments in tourism forecasting practice: How good are they? Tourism Economics, 25(3), 402–424. DOI: 10.1177/1354816618806727. https://doi.org/10.1177/1354816618806727
  6. Lin, V. S., & Song, H. (2015). A review of Delphi forecasting research in tourism. Current Issues in Tourism, 18(12), 1099–1131. DOI: 10.1080/13683500.2014.967187. https://doi.org/10.1080/13683500.2014.967187
  7. Arvan, M., Fahimnia, B., Reisi, M., & Siemsen, E. (2019). Integrating human judgement into quantitative forecasting methods: A review. Omega, 86, 237–252. DOI: 10.1016/j.omega.2018.07.012. https://doi.org/10.1016/j.omega.2018.07.012
  8. van Dijk, D., & Franses, P. H. (2019). Combining expert-adjusted forecasts. Journal of Forecasting, 38, 415–421. DOI: 10.1002/for.2570. https://doi.org/10.1002/for.2570
  9. Song, H., Qiu, R. T. R., & Park, J. (2019). A review of research on tourism demand forecasting: Launching the Annals of Tourism Research Curated Collection on tourism demand forecasting. Annals of Tourism Research, 75, 338–362. DOI: 10.1016/j.annals.2018.12.001. https://doi.org/10.1016/j.annals.2018.12.001
  10. Athanasopoulos, G., Hyndman, R. J., Kourentzes, N., & O'Hara-Wild, M. (2023). Probabilistic forecasts using expert judgment: The road to recovery from COVID-19. Journal of Travel Research, 62(1), 233–258. DOI: 10.1177/00472875211059240. https://doi.org/10.1177/00472875211059240
  11. Koupriouchina, L., van der Rest, J.-P., & Schwartz, Z. (2023). Judgmental adjustments of algorithmic hotel occupancy forecasts: Does user override frequency impact accuracy at different time horizons? Tourism Economics, 29(8), 2143–2164. DOI: 10.1177/13548166221126572. https://doi.org/10.1177/13548166221126572
  12. Wu, X., Shi, J., & Xiong, H. (2024). Tourism forecasting research: a bibliometric visualization review (1999–2022). Tourism Review, 79(2), 465–486. DOI: 10.1108/TR-03-2023-0169. https://doi.org/10.1108/TR-03-2023-0169
  13. Song, H., & Zhang, H. (2025). Tourism demand modelling and forecasting: a Horizon 2050 paper. Tourism Review, 80(1), 8–27. DOI: 10.1108/TR-12-2023-0882. https://doi.org/10.1108/TR-12-2023-0882
  14. Li, F., & Ruan, T. (2026). Recovery-informed forecasting strategy enhancement. Annals of Tourism Research, 118, 104164. DOI: 10.1016/j.annals.2026.104164. https://doi.org/10.1016/j.annals.2026.104164
  15. Wang, H., Gong, Y., Pan, B., & Hu, T. (2026). Good enough tourism demand forecasting: Insights from local practice in China. Annals of Tourism Research Empirical Insights, 7, 100225. DOI: 10.1016/j.annale.2026.100225. https://doi.org/10.1016/j.annale.2026.100225