From Tables to Time: How TabPFN-v2 Outperforms Specialized Time Series Forecasting Models

TL;DR

TabPFN-TS demonstrates that a compact, tabular-foundation-model approach can achieve state-of-the-art-like performance on time series forecasting without time-series-specific pretraining. By converting series into a tabular regression problem and enriching inputs with calendar, automatically discovered seasonalities, and a running index, the method delivers competitive point forecasts and superior probabilistic forecasts on GIFT-Eval. The work highlights the generality and potential of tabular foundation models for time series tasks, while also identifying practical bottlenecks like inference speed and opportunities for further enhancement through pretraining, fine-tuning, and covariate integration. Overall, it suggests a promising new direction where tabular priors and feature engineering can rival specialized time-series architectures.

Abstract

Foundation models have become increasingly popular for forecasting due to their ability to provide predictions without requiring a lot of training data. In this work, we demonstrate how TabPFN-v2, a general tabular foundation model, can be effectively applied to time series forecasting. We introduce TabPFN-TS, a simple method that combines TabPFN-v2 with lightweight feature engineering to enable both point and probabilistic forecasting. Despite its simplicity and compact size (11M parameters), TabPFN-TS achieves top rank on the public GIFT-Eval leaderboard in both forecasting tasks. Through ablation studies, we investigate factors contributing to this surprising effectiveness, especially considering TabPFN-v2 was pretrained solely on synthetic tabular data with no exposure to time series. Our results highlights the potential of tabular foundation models like TabPFN-v2 as a valuable new approach for time series forecasting. Our implementation is available at https://github.com/PriorLabs/tabpfn-time-series.

Figures (15)

• Figure 1: Overview of TabPFN-TS. Given a time series, we derive features from the timestamps to form both X_train and X_test. The target values of the history are used as y_train. These three variables are then used by TabPFN to predict the target values of the future timestamps.
• Figure 2: Forecasting performance of TabPFN-TS and baseline models on all 97 GIFT-Eval benchmarking tasks. TabPFN-TS ranks #1 in probabilistic forecasting (WQL, both raw and rank) and #2 in point forecasting (MASE). WQL and MASE are normalized by Seasonal Naive, and aggregated by geometric mean. Model ranks are aggregated by arithmetric mean. Error bars indicate $95\%$ confidence intervals.
• Figure 3: The performance of TabPFN-TS is significantly influenced by the selected time series featurization. In this analysis, we demonstrate that each component of our featurization enhances performance. Additionally, we highlight the substantial difference between encoding a time series in the simplest form (using only the index feature) and utilizing the featurization we propose.
• Figure 4: Predictions of TabPFN-v2 on $\sin(nx)$ for $n = 1,\dots,24$. (a) With only $\sin(x)$ as input feature, the model accurately approximates $\sin(nx)$ for odd values of $n$. (b) When given both $\sin(x)$ and $\cos(x)$, it successfully approximate $\sin(nx)$ for both odd and even integers $n$.
• Figure 5: Predictions of TabPFN-v2 on composite sinusoidal signals when given $\sin(x)$ and $\cos(x)$ as input features. Each composite signal is the sum of 3--10 sinusoids with randomly chosen frequencies $f \in [1, 24]$, amplitudes $A \in [0.5, 2.0]$, and phase shifts $\phi \in [0, 2\pi]$.