Verified, then logged.
Sun 22 Nov 2026 · 00:20 KST / 15:20 UTC한국어 | ENSearch /

04 / Policy · Analysis

GPT study forecasts Korean consumption coupon effects

TechM reports a reduction in forecasting error, but gives no concrete procedure for incorporating the findings into policy design.

Sample vials, a well plate and a pipette sit on a biosafety lab bench before a monitor showing a protein structure model, with a researcher blurred in the background, seen from behind. AI-generated image

Kwon Hyuk-gu, a chair professor at Singapore’s Nanyang Technological University, presented research using GPT to forecast policy effects, Korean technology publication TechM reported on 11 Nov. Supported by OpenAI, the study tests forecasts of the spending stimulus from Korea’s consumption coupons.

The result described by TechM concerns forecasting accuracy. The reported reduction in prediction error measures the experiment’s performance in estimating the coupons’ effect on spending. The report gives no concrete procedure for incorporating the findings into policy design.

According to TechM, the experiment examines consumption coupons introduced in Korea last year to support livelihoods and economic recovery. Applying GPT to data on this domestic programme makes the study a case of AI-assisted policy evaluation in Korea.

Forecasting the coupons’ effect on spending

TechM reported that Kwon, a chair professor in the university’s information systems department, attended an OpenAI media session at the company’s office in Gangnam, a district of Seoul. He presented the research at that session.

According to TechM, the researchers used GPT to analyse data with OpenAI’s support. Their experiment focused on predicting the consumption coupons’ effect on spending.

The scope of the forecasting improvement

TechM reported that the prediction error fell from 25% early in the research to around 10%. Those figures describe the study’s forecasting error, rather than the policy’s budget or the rate of growth in consumption.

The report presents an improvement in a forecasting experiment, according to TechM’s account. Applying that result to policy design remains a separate step for which the report provides no concrete procedure.