Predicting recessions using AI
In economics, efforts to predict the economic cycle and especially recessions are central. The ability to respond early to economic turmoil can mean the difference between success and failure for companies, investors and governments. In recent years, artificial intelligence (AI) has taken on an impressive role in improving our ability to predict economic downturns. Can AI be considered a reliable tool for identifying recessions?
Impending recession in the USA?
The discussion about whether the USA will be hit by a recession in 2024 is in full swing. Goldman Sachs estimates the probability at 15%. The New York Federal Reserve Bank, on the other hand, forecasts 69% using a yield curve, a popular indicator for determining recessions. Surveys of CEOs and consumers show that 84% and 69%, respectively, believe a recession is likely in the next 12-18 months. Quantitative analysts, also known as “quants,” are now asking whether artificial intelligence can be used to improve accuracy. However, it has been shown that machine learning is not as useful as initially hoped.
Weaknesses of AI
“Recession modeling has not evolved as much compared to other areas,” says Max Gokhman, head of investment strategy MosaiQ at Franklin Templeton. This is due to two main reasons. On the one hand, there is simply not enough data available to train such models effectively. On the other hand, it is difficult to isolate the relevant economic signals and evaluate them appropriately. Data is the foundation of artificial intelligence, but is limited in this context. Since the 1990s, there have only been four recessions in the U.S. based on GDP indicators. Therefore, the learning ability of the models is severely limited, which significantly hinders their complexity. Aric Whitewood, co-founder of AI macro hedge fund to effectively train a model.
Attempt from BNPPAM
Quants divide recession forecasting into two main areas: identifying variables that indicate an impending recession and finding innovative statistical techniques to develop fresh data sources for forecasting. Similar to economists, they rely on various indicators to determine recessions, including Treasury yield curve inversion, fluctuations in gross domestic product (GDP), or Markov switching models. In addition, they use dynamic factor models, which assume that the common movement in a large number of time series can be explained by a limited number of unobserved common factors. One of these latent factors is the business cycle. Asset management firm BNPPAM has developed models based on macroeconomic variables to create an early warning indicator of recessions. Unfortunately, the resulting signals were too weak and they could not be used successfully in real-time trading. Due to the renewed lack of sufficient data and the enormous complexity of the financial markets, the project has been put on hold indefinitely.
Deep Learning
Meanwhile, Franklin Templeton has been working extensively on deep learning to further analyze the network of factors that contribute to recessions. Deep learning represents a branch of machine learning that aims to imitate the functioning of the human brain. In this method, the machine learning process searches for patterns in historical data by changing the coefficients of formulas defined by quant analysts. The relationship to predicted outcomes evolves over time. The challenge here is that the model recreated the formulas from scratch, resulting in the data becoming opaque and unusable. In conclusion, it can be said that AI cannot (yet) be used as a reliable warning system for recessions. This situation is undoubtedly likely to change in the coming years as models and overall technology continue to be refined.
IT Jobs
Administration Jobs
Beauty Jobs
Education Jobs
Finance Jobs
Healthcare Jobs
Human Resources Jobs
Logistics Jobs 
Pharma Jobs
Jetzt teilen
Jetzt teilen