Monex Securities presents a practical approach to forecasting the Nikkei Stock Average weekly using ChatGPT. The article explains the procedure for calculating the expected weekend closing price and expected price change based on the previous Friday's closing price, and examines how to use the variability in ChatGPT's responses as an indicator of market sentiment.
As a practical application of AI forecasting, Monex Securities uses the familiar ChatGPT to forecast the closing price and price change for the end of each week. Every Monday at 8:00 a.m., it provides ChatGPT with the previous Friday's closing price and asks it to produce a single forecast deemed most likely, based on information researched from the perspective of a strategist specializing in the stock market, while considering both upward and downward factors.
In the latest forecast exercise, 10 forecasts conducted on 19 October 2026 produced an average closing price of ¥65,210, approximately ¥114 below the previous weekend's close. At the same time, only 5 of the forecasts called for an increase, indicating variability in the responses. The article examines whether this variability can be used to assess not only the stock price level at the end of the week, but also market sentiment.
In an 8-week test, the direction of the expected closing price matched the actual weekend result in 7 weeks, for an accuracy rate of 87.5%, while the accuracy rate for market sentiment was 85.7%. By contrast, a simulation covering the past 41 weeks produced a direction match rate of 56.1% for the expected closing price and a market sentiment match rate of 52.4%, with some metrics declining over the longer testing period.
Rather than using AI forecasts directly to make investment decisions, the article proposes using the average forecast value derived from multiple responses together with their variability as an indicator of market sentiment. Through a review of the results and continued testing, it presents a practical approach that positions AI as an analytical tool.