Bessey searches for new models to reduce prediction errors

2024-08-26 18:36:15

Exterior of the Ministry of Economy and Finance in Paris, March 20, 2024. Emmanuel Dunand/AFP

Forecasting is definitely a difficult art, especially when it comes to the future, and even more so in times of crisis. Economists at Bessie know this. Since the Covid-19 pandemic and the war in Ukraine, the economic indicators they rely on, especially for preparing budgets, have lost some of their relevance. Forecast errors are getting worse, and the search for new, more reliable models has become both important and difficult. Here are three experts explaining it An unusual note issued by the General Administration of Finance on August 22.

Unusually, statisticians rarely look back on their own or their colleagues’ failures. But playing ostrich has proved impossible here, because the gap between growth forecasts and reality has been so wide during a health crisis. The report’s authors, Lina Bourassi, Antoine Claisse and Louise Phung, compare the recession in early 2009 with the one caused by the first lockdown 11 years later. The deterioration in the “business environment” shown in the 2020 survey was 1.2 times greater than in 2009, according to the Banque de France, 1.9 times more, according to INSEE and 2.6 times more, according to S&P. In fact, however, the drop in GDP was almost six times as large. What a turnaround.

What explains this? The note cites several reasons. One is the types of questions asked of companies in these surveys. “How has your production been going over the past month? » Increase, decrease, stability: the responses aggregated by INSEE do not indicate the extent of the changes. First, the usual composite indicators combine data reflecting the demand for products and services. However, with Covid-19 and the war in Ukraine, the evolution of the French economy “Mainly affected by supply factors” The Ministry of Finance pointed out that, for example, supply and recruitment difficulties. What’s more, the increase in variables such as delivery times or raw material prices, which traditionally reflected strong demand and brisk activity, suddenly became a reflection of supply difficulties, leading to slower production and slower growth. Enough to interfere with the reading of the numbers.

Alternative Models

Bercy launched a complete overhaul of its predictive model in late 2022. The new device is at least using artificial intelligence, if not artificial intelligence. Automatic variable selectionhas been developed. Millions of data combinations were tested, incorporating factors such as the number of companies affected by employee shortages, stock market indices, and the prices of gold and the euro. The results? It looks pretty amazing. These alternative models show performance “Much better” The Treasury Department noted that compared with traditional tools, the error rate was almost divided by three.

16.83% of this article remains to be read. The rest is reserved for subscribers.

1724712744
#Bessey #searches #models #reduce #prediction #errors