Illustration comparing MIDAS and Lag-Llama models in inflation forecasting

Inflation Nowcasting: Can AI Beat Traditional Models?

Rhea Morgan in Business & Economy December 2025 • 4 min read.

"A comparative look at how Mixed-Data Sampling (MIDAS) and Lag-Llama models forecast inflation, and what it means for economic predictions."

Inflation is a critical economic indicator that everyone from public institutions to private investors closely monitors. Accurate inflation forecasts help policymakers make informed decisions, assist businesses in planning, and guide individuals in making financial choices. The ability to predict inflation trends accurately can provide a significant advantage in navigating economic landscapes.

Traditionally, economists have relied on econometric models to forecast inflation. However, recent advancements in artificial intelligence (AI) have introduced new tools that might offer even better predictions. One such AI model is Lag-Llama, a foundational time series forecasting model based on a Long Short-Term Memory (LSTM) neural network. This model has shown promise in various forecasting tasks, raising the question: Can AI outperform traditional methods in predicting inflation?

A recent study compared the performance of a traditional econometric model, Mixed Data Sampling (MIDAS), with Lag-Llama in nowcasting the Harmonized Index of Consumer Prices (HICP) in the Euro area. This analysis sought to determine whether Lag-Llama could outperform MIDAS, a well-established model, under optimal conditions. By evaluating both models across several metrics, the study sheds light on the potential of AI in economic forecasting.

MIDAS vs. Lag-Llama: Decoding the Models

Illustration comparing MIDAS and Lag-Llama models in inflation forecasting

To understand the study's findings, it's essential to grasp the basics of the two models being compared:

Mixed Data Sampling (MIDAS): This model, developed by Ghysels et al., is designed to handle data sampled at different frequencies. Unlike traditional models that require all data to be at the same frequency, MIDAS can incorporate high-frequency data (like daily financial data) to predict lower-frequency data (like monthly inflation rates) without needing aggregation. This capability is particularly useful in economic forecasting, where timely data can provide valuable insights.

Handles mixed-frequency data without aggregation.
Enhances forecasting accuracy by incorporating real-time data.
Widely adopted by public institutions and private entities.

Lag-Llama: A foundational time series forecasting model based on Long Short-Term Memory (LSTM) neural networks. Developed by Rasul et al., Lag-Llama is trained on a large number of time-series datasets and can perform zero-shot forecasting, making accurate predictions on datasets it has never been explicitly trained on. It uses a distribution head, projecting features to parameters of a probability distribution, offering probabilistic forecasting, a departure from traditional point estimates.

The Verdict: Can AI Conquer Inflation Forecasting?

The study revealed that Lag-Llama, the AI-driven model, generally outperformed the traditional MIDAS regression in forecasting Eurozone HICP. Lag-Llama achieved better scores across key metrics such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Mean Squared Error (MSE). It also demonstrated a higher correlation with actual inflation data. These results suggest that AI has the potential to enhance the accuracy of economic forecasting, providing more reliable insights for decision-makers.

About this Article -

This article was crafted using a human-AI hybrid and collaborative approach. AI assisted our team with initial drafting, research insights, identifying key questions, and image generation. Our human editors guided topic selection, defined the angle, structured the content, ensured factual accuracy and relevance, refined the tone, and conducted thorough editing to deliver helpful, high-quality information.See our About page for more information.

This article is based on research published under:

DOI-LINK: https://doi.org/10.48550/arXiv.2407.0851,

Title: Comparative Analysis Of Mixed-Data Sampling (Midas) Model Compared To Lag-Llama Model For Inflation Nowcasting

Subject: econ.em

Authors: Adam Bahelka, Harmen De Weerd

Published: 11-07-2024

Everything You Need To Know

What is the key difference between Mixed Data Sampling (MIDAS) and Lag-Llama in terms of data handling?

The primary distinction lies in how they handle data frequencies. MIDAS is specifically designed to work with mixed-frequency data, meaning it can incorporate data sampled at different intervals (e.g., daily and monthly) without needing to aggregate it. This allows it to leverage high-frequency data to predict lower-frequency data more effectively. Lag-Llama, on the other hand, is a time series forecasting model that can handle various data frequencies through its LSTM neural network architecture, which is trained on a large number of time-series datasets and can perform zero-shot forecasting.

How does Lag-Llama's architecture contribute to its forecasting capabilities, and what advantages does it offer over traditional methods like MIDAS?

Lag-Llama, built on a Long Short-Term Memory (LSTM) neural network, offers several advantages. Its architecture allows it to be trained on extensive time-series datasets, enabling it to perform zero-shot forecasting, meaning it can accurately predict on datasets it has not been explicitly trained on. This contrasts with traditional models like MIDAS, which rely on econometric principles and may require more specific data preparation and model tuning. Furthermore, Lag-Llama uses a distribution head, providing probabilistic forecasting rather than just point estimates, offering a more comprehensive view of potential outcomes.

What are the key metrics used to compare the performance of MIDAS and Lag-Llama, and what do they indicate about their forecasting accuracy?

The study assessing the performance of MIDAS and Lag-Llama utilized metrics such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Mean Squared Error (MSE). These metrics quantify the magnitude of the errors in the forecasts. Additionally, the study considered the correlation between the model's predictions and actual inflation data. Better scores across these metrics, along with a higher correlation, indicate superior forecasting accuracy. For example, a lower MAE, MAPE, and MSE suggest that the model's predictions are closer to the actual values, implying higher accuracy. The study found that Lag-Llama generally outperformed MIDAS across these metrics.

In the context of inflation forecasting, why is the ability to handle mixed-frequency data an important feature, as seen in the MIDAS model?

The ability to handle mixed-frequency data is crucial in inflation forecasting because it allows models to incorporate a wider range of timely data. Economic indicators are often released at different intervals; for instance, financial data may be available daily, while inflation rates are reported monthly. MIDAS can incorporate these high-frequency data points to enhance the accuracy of predictions. By using real-time data, MIDAS can provide a more accurate and up-to-date assessment of inflation trends, which is valuable for making informed economic decisions.

How could the enhanced accuracy of AI models like Lag-Llama impact financial forecasting and decision-making for both public institutions and private investors?

The improved accuracy of AI models such as Lag-Llama can significantly impact financial forecasting and decision-making. For public institutions, more reliable inflation forecasts can inform monetary policy decisions, helping to manage inflation effectively and stabilize the economy. Accurate predictions enable policymakers to anticipate economic changes and adjust interest rates or other measures proactively. Private investors can use these improved forecasts to make more informed investment decisions, allocate assets efficiently, and manage risk more effectively. By having better insights into inflation trends, both sectors can make more strategic and informed decisions, leading to potentially higher returns and economic stability.