This article puts AI-based forecasts for the Nikkei Stock Average to the test. Using ChatGPT, it makes forecasts for one day, one week and two weeks ahead, examining how repeated questions produce fluctuations in forecast values and multiple scenarios.
Generative AI has increasingly been applied to share-price forecasting in recent years. When prompted, it may provide not only a projected direction but also a specific forecast value and the reasoning behind it. The article suggests using AI forecasts as input based on information available at the time and the prevailing market environment, rather than accepting them at face value.
It first points out that AI share-price forecasts have limitations, including the possibility that new information unknown at the time of the forecast may emerge later and move share prices. The longer the forecast horizon, the more information can emerge and the more forecasts are likely to diverge, making it important to consider the information cut-off and forecast period.
In an experiment asking ChatGPT to forecast the Nikkei Stock Average’s trading range and closing level, repeating the same question produced variations in the forecast values. Even when the average of 10 runs pointed to an upward direction, there were cases in which the index actually declined. The article explains that comparing multiple forecasts and assessing their consistency can help make use of this variability.
In forecasts for the Nikkei Stock Average two weeks ahead, most of the repeated responses often pointed to an increase. However, opposing factors such as recession concerns and yen appreciation were sometimes cited in just 1 response, highlighting the importance of considering both a base case and a risk scenario.
The article concludes that the key to applying AI to investment management is not simply to have it provide “the answer,” but to use multiple responses as input and add one’s own interpretation.