TimeRSI: Towards Self-Improved Agentic Deployment for Time-Series Foundation Models

Haoxin Liu; Deqing Fu; Lucheng Fu; Yiqiao Jin; Rushi Qiang; Harshavardhan Kamarthi; B. Aditya Prakash

cs.LGOct 14, 2026

TimeRSI: Towards Self-Improved Agentic Deployment for Time-Series Foundation Models

Authors: Haoxin Liu, Deqing Fu, Lucheng Fu, Yiqiao Jin, Rushi Qiang, Harshavardhan Kamarthi, B. Aditya Prakash

Organizations: Georgia Institute of Technology · University of Southern California · Samay AI

Abstract

Time-series foundation models provide strong zero-shot forecast, but deployment for real-world applications still requires task-specific designs for model inputs and outputs. Direct inference can leave domain knowledge and recurring forecast errors unaddressed, while tailored pipelines require substantial human expertise and effort. We introduce TimeRSI, a self-improving agentic deployment system around frozen forecasting models. A coding agent translates task descriptions into executable pipeline changes. It tests each revision on historical forecasting windows and uses observed errors to guide subsequent revisions. During deployment, an adaptive selector chooses among the resulting pipelines using completed forecasts under similar conditions. Across 198 tasks on TIME and fev-bench, TimeRSI outperforms all submissions on six point and probabilistic forecasting metrics. Controlled comparisons support the value of feedback-guided revision over non-iterative pipeline search. Further experiments verify the generalization with different coding agents and forecasting models. Code is available at https://github.com/haoxin1998/TimeRSI.

Figures & tables

Panel of Figure 1 — Fig. 1: From conventional to self-improved TSFM deployment. (a) Direct inference leaves task-specific adaptation to the user, while tailored pipelines require human design and revision. (b) TimeRSI automates pipeline improvement through task-aware code edits and historical forecasting feedback, then selects among the resulting pipelines for each new forecast. The TSFM remains frozen.

Fig. 2: TimeRSI ranks first among the compared submissions across six reported metrics on TIME and FEV-bench . Labels name the frozen backbone of TimeRSI ; LPTM ( Kamarthi and Prakash, 2024 ) is appended below the dashed line when outside the top ten. See App. E for metrics including WQL Skill and WAPE Skill, and full results of more methods.

Panel of Figure 3 — Fig. 3: Agentic deployment examples via TimeRSI . The agent adds calendar features for parking, separates tidal patterns for ocean currents, applies holiday rules for supply-chain demand, and restores daily accumulation and reset patterns for solar counters. These agentic deployments reduce MASE by 2.86–6.70% over direct TimesFM 3.0.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

		Test MASE $\downarrow$		Relative error reduction
Task	Horizon	Direct	TimeRSI	MASE	CRPS
Supply chain	30 days	0.557750	0.520378	6.70%	6.76%
Parking availability	1 week	0.678730	0.641658	5.46%	5.37%
Ocean currents	72 hours	0.871531	0.835129	4.18%	0.64%
Solar counters	72 hours	0.574087	0.557693	2.86%	2.66%
Auto-production seasonal factor	12 months	0.756537	0.233980	69.07%	66.37%

Appendix

Tab. 1: Different tasks lead to different deployment programs. Complete-task test scores for the selected examples. Relative reductions compare TimeRSI with direct TimesFM inference. The auto-production case uses an agent-written specialist for point forecasts and the foundation model for distributional information.

Revision	MASE $\downarrow$
Starting pipeline	0.522271
1. Holiday and weekday corrections	0.489970
2. Post-holiday adjustment	0.490036
3. Align related forecast shapes	0.489901
4. Stronger post-holiday adjustment	0.489933
5. Restrict adjustment to July 5	0.489834

Appendix

Tab. 2: Supply-chain revisions evaluated on the same adaptation windows. The starting pipeline already contains earlier adaptations; its score is not the direct-inference baseline.

Revision	Adaptation	Selection
Direct TimesFM	0.9421	0.6142
1. Calendar-expert blend	0.9578	0.6171
2. Calendar inputs	0.9208	0.6038
3. History-length condition	0.9208	0.6038
4. Weekend–time features	0.9268	0.6010
5. Median and range edits	0.9208	0.6038

Appendix

Tab. 3: Parking MASE on adaptation and separate selection windows; lower is better. Adaptation uses three origins and selection uses five, each with 32 car parks. Comparisons are within columns. Revision 4 is deployed, rather than the last revision or the adaptation winner.

Revision	Adaptation	Selection
Direct TimesFM	3.9173	5.0403
1. Log transformation	4.1816	4.2079
2. Damped trend blend	2.9624	3.8822
3. Repeated-value handling	2.7920	4.1122
4. Shorter context	2.7604	4.1589
5. Growth-duration damping	2.7252	4.1114

Appendix

Tab. 4: COVID-report MASE on separate historical sets. Selection favors revision 6, which uses milder damping, over revision 5, which has the lowest adaptation error.

Method	MASE $\downarrow$	CRPS $\downarrow$
TimeRSI	0.630	0.528
QiYao-M	0.638	0.536
ChakraTS	0.639	0.537
TimesFM 3.0	0.640	0.536
Toto 2.0 (2.5B)	0.642	0.539
Toto 2.0 (313M)	0.644	0.542

Appendix

Tab. 5: Full public-result comparison. All visible methods in the captured snapshot. Bold marks TimeRSI . TIME values are normalized errors; FEV-bench values are Skill Scores. TIME includes the accepted entry; the FEV-bench TimeRSI row is the submitted, not yet merged result. Rows are ordered by TIME MASE and FEV-bench SQL Skill.

TIME : normalized error $\downarrow$
Metric	Direct	TimeRSI	Gain
MASE	0.639759	0.630011	1.52%
CRPS	0.536328	0.527713	1.61%
FEV-bench : Skill Score (%) $\uparrow$
MASE	37.786	40.579	4.49%
SQL	48.984	51.243	4.43%

Appendix

Tab. 6: Paired comparison with direct inference. The reference uses our environment and the same backbone. Gain denotes relative error reduction. For FEV-bench it is computed from

100-\text{Skill}

. The public FEV TimesFM row differs slightly from this local reference.

Configuration	MASE $\downarrow$	CRPS $\downarrow$
TimesFM 3.0 backbone
Direct inference	0.640	0.536
TimeRSI : Claude Code	0.630	0.528
TimeRSI : Antigravity	0.637	0.532
Chronos-2 backbone
Direct inference	0.662	0.556

Appendix

Tab. 7: Deployment gains across coding agents and TSFMs. Normalized errors on all 98 TIME tasks. Claude Code uses Claude Fable 5; Antigravity uses Gemini 3.8 Flash. Bold marks the best scores within each backbone group.

Method	SQL Skill $\uparrow$	MASE Skill $\uparrow$	WQL Skill $\uparrow$	WAPE Skill $\uparrow$
TimeRSI	51.243	40.579	55.759	44.788
TimesFM 3.0	48.994	37.816	53.278	41.656
ChakraTS	48.143	36.930	52.440	40.576
Chronos-2	47.281	35.496	51.520	39.410
Lingjiang2.0	47.020	35.377	51.457	39.229
T0-beta	46.653	34.863	50.827	38.318

Appendix

Tab. 8: Full FEV-bench comparison across four metrics. Skill Scores (%; higher is better) cover all 100 tasks. We include all 31 publicly visible baselines and our submitted result. Rows are ordered by SQL Skill; bold marks the best score in each column.

Agentic Time Series Forecasting

See all 14 papers

When Tomorrow Becomes Today: Self-Evolving Policies for Agentic Time-Series Forecasting
Sep 21, 2026Yifan Hu, Xilin Dai, Zhiyuan Qu +4Online Time Series Forecasting Time Series Forecasting
Ant International · Tsinghua University · The Chinese University of Hong Kong
Expert-Guided Forecast Editing for Time-Series Foundation Models
Jul 22, 2026Hung Le, Minh Hoang Nguyen, Manh Nguyen +2Black-Box Optimization Time Series Forecasting
Deakin Applied Artifical Intelligence Initiative Deakin University, Australia
TimeRouter: Efficient and Adaptive Routing of Time-Series Foundation Models
Jun 10, 2026Kanghui Ning, Yushan Jiang, Kashif Rasul +3Adaptive Model Routing Time Series Forecasting
School of Computing, University of Connecticut, Storrs, USA · Department of Machine Learning Research, Morgan Stanley, New York, USA.

Sep 21, 2026J/K move · Enter open · S save

When Tomorrow Becomes Today: Self-Evolving Policies for Agentic Time-Series Forecasting

Yifan Hu, Xilin Dai, Zhiyuan Qu, Yiding Liu+3

Ant International · Tsinghua University · The Chinese University of Hong Kong

Online Time Series Forecasting Time Series Forecasting Agentic Time Series Forecasting Self-Evolving Agents

Open paper AskPDF arXiv

Agentic time series forecasting concerns systems whose underlying mechanisms evolve, making the relative effectiveness of numerical models, reasoning strategies, and intervention rules inherently time-varying. Consequently, a time series agent must adapt the forecasts it produces and the orchestration policy that determines which components to trust and how to coordinate them. The deployment process naturally provides supervision for this adaptation as forecast horizons elapse and realized targets reveal the effectiveness of earlier decisions. Committing all numerical expert forecasts and candidate agent paths before target observation allows each realized outcome to evaluate the entire alternative set, providing delayed feedback without additional annotation. However, existing time series agents primarily incorporate prior experience through forecast refinement, reflection, or retrieval, without systematically converting realized outcomes into persistent updates to the joint orchestration policy governing later origins. To exploit this delayed feedback systematically, we introduce TimEvolve, a frozen-backbone time series agent that converts each realized outcome into persistent joint updates of expert trust, agent path selection, and intervention strength. A temporally ordered predict, reveal, and update protocol applies this feedback to subsequent forecasts. Experiments across eight Time-MMD domains show that TimEvolve achieves the best average MSE and MAE ranks among fifteen methods and the lowest errors on both metrics in seven domains. These results demonstrate the value of learning forecasting policies from the futures encountered during deployment.

Abstract

Agentic Time Series Forecasting

Explore similar work

When Tomorrow Becomes Today: Self-Evolving Policies for Agentic Time-Series Forecasting

Expert-Guided Forecast Editing for Time-Series Foundation Models

TimeRouter: Efficient and Adaptive Routing of Time-Series Foundation Models