--- title: Forking-Sequences — Part II: Multi-Horizon Forecast... description: This CMU ML blog post (Part II of a series) presents forking-sequences as a mechanism for multi-horizon forecast ensembling with reduced volatility. Key... canonical: https://daily.dev/posts/forking-sequences-part-ii-multi-horizon-forecast-ensembling-with-reduced-volatility-lp5meah1m twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Forking-Sequences — Part II: Multi-Horizon Forecast Ensembling with Reduced Volatility | daily.dev og:description: This CMU ML blog post (Part II of a series) presents forking-sequences as a mechanism for multi-horizon forecast ensembling with reduced volatility. Key... og:url: https://daily.dev/posts/forking-sequences-part-ii-multi-horizon-forecast-ensembling-with-reduced-volatility-lp5meah1m og:image: https://api.daily.dev/og/posts/lp5meAh1M.png og:image:alt: Forking-Sequences — Part II: Multi-Horizon Forecast Ensembling with Reduced Volatility og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Forking-Sequences — Part II: Multi-Horizon Forecast Ensembling with Reduced Volatility **[ML CMU](https://daily.dev/sources/mlcmu)** · 9 min read · 1 upvotes · 0 comments ## Summary This CMU ML blog post (Part II of a series) presents forking-sequences as a mechanism for multi-horizon forecast ensembling with reduced volatility. Key contributions include two new forecast volatility metrics — scaled Forecast Percentage Change (sFPC), usable in real-time without ground truth, and scaled Excess Volatility (sEV), which rewards accuracy-improving revisions and only penalizes harmful ones. Empirically, applying exponential-smoothing ensembling (α=0.9) to forking-sequences models reduces sEV by 10–13% across RNN, LSTM, CNN, Transformer, and StateSpace architectures with less than 0.1% accuracy degradation. The technique also works zero-shot on pretrained Time Series Foundation Models (Chronos-2, Toto 2.0, TimesFM, PatchTST, N-BEATS), achieving ~10% volatility reduction at negligible accuracy cost. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is scaled Excess Volatility (sEV) and how is it different from just measuring forecast changes? Scaled Excess Volatility (sEV) is a probabilistic forecast volatility metric that only penalizes revisions moving a forecast away from ground truth or overshooting it, unlike naive percentage-change metrics that treat every revision as equally bad. It assigns zero penalty to accuracy-improving revisions, full quantile-loss penalty to deteriorating ones, and a partial penalty for overshoots, distinguishing beneficial revisions from harmful volatility. _Teams weighing new volatility metrics for forecasting pipelines can track this research area on daily.dev._ ### How much does exponential smoothing ensembling reduce forecast volatility for forking-sequences models without hurting accuracy? Applying exponential smoothing at inference with alpha=0.9 reduces scaled Excess Volatility (sEV) by 13.2%, 13.0%, 10.9%, 10.2%, and 11.2% for RNN, LSTM, CNN, Transformer, and StateSpace-based encoders respectively, while degrading forecast accuracy (sCRPS) by less than 0.1%. Lower alpha values cut volatility further but at greater accuracy cost. _Engineers tuning forecast ensembling trade-offs can follow developments like this on daily.dev._ ### Can forecast ensembling reduce volatility on pretrained time series foundation models like Chronos-2 or TimesFM without retraining? Yes, forecast ensembling can be applied zero-shot to pretrained models originally trained with window-sampling, including Chronos-2, Toto 2.0, TimesFM, PatchTST, and N-BEATS, by aggregating their forecast revision outputs. Across the M-series benchmark this achieved a median roughly 10% reduction in forecast volatility with less than 0.1% degradation in accuracy (sCRPS). _Practitioners evaluating foundation models for forecasting can keep up with such techniques via daily.dev._ ## Similar posts on daily.dev - [Forking-Sequences — Part I: Statistically and Computationally Efficient Multi-Horizon Forecasting](https://daily.dev/posts/forking-sequences-part-i-statistically-and-computationally-efficient-multi-horizon-forecasting-0sj00ydhh) · ML CMU · 3 upvotes · 0 comments - [Five Questions About Chronos-2, the Time Series Foundation Model](https://daily.dev/posts/five-questions-about-chronos-2-the-time-series-foundation-model-owv5sq0op) · Towards Data Science · 0 upvotes · 0 comments - [Toto 2.0: Time series forecasting enters the scaling era](https://daily.dev/posts/toto-2-0-time-series-forecasting-enters-the-scaling-era-t2iz3dkjh) · Datadog · 3 upvotes · 0 comments - [The 2026 Time Series Toolkit: 5 Foundation Models for Autonomous Forecasting](https://daily.dev/posts/the-2026-time-series-toolkit-5-foundation-models-for-autonomous-forecasting-gkjr71x9v) · Machine Learning Mastery · 4 upvotes · 0 comments - [The Unreasonable Difficulty of Time Series Forecasting](https://daily.dev/posts/the-unreasonable-difficulty-of-time-series-forecasting-yozpddbsw) · Hacker News · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#deep-learning](https://daily.dev/tags/deep-learning), [#neural-networks](https://daily.dev/tags/neural-networks), [#time-series-forecasting](https://daily.dev/tags/time-series-forecasting) [View this post on daily.dev](https://daily.dev/posts/forking-sequences-part-ii-multi-horizon-forecast-ensembling-with-reduced-volatility-lp5meah1m) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Forking-Sequences — Part II: Multi-Horizon Forecast Ensembling with Reduced Volatility","url":"https://daily.dev/posts/forking-sequences-part-ii-multi-horizon-forecast-ensembling-with-reduced-volatility-lp5meah1m","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/forking-sequences-part-ii-multi-horizon-forecast-ensembling-with-reduced-volatility-lp5meah1m"},"datePublished":"2026-08-10T22:01:31.848Z","dateModified":"2026-09-13T21:58:04.643Z","description":"This CMU ML blog post (Part II of a series) presents forking-sequences as a mechanism for multi-horizon forecast ensembling with reduced volatility. Key...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/101d9823bf171d51265a1ce47b2a9651?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/101d9823bf171d51265a1ce47b2a9651?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"ML CMU","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"ML CMU","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/f6a0448c0a90422d9a19c706a926c0ad","url":"https://daily.dev/sources/mlcmu"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/forking-sequences-part-ii-multi-horizon-forecast-ensembling-with-reduced-volatility-lp5meah1m","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,deep-learning,neural-networks,time-series-forecasting","timeRequired":"PT9M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"ML CMU","item":"https://daily.dev/sources/mlcmu"},{"@type":"ListItem","position":3,"name":"Forking-Sequences — Part II: Multi-Horizon Forecast Ensembling with Reduced Volatility"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/forking-sequences-part-ii-multi-horizon-forecast-ensembling-with-reduced-volatility-lp5meah1m#faq","mainEntity":[{"@type":"Question","name":"What is scaled Excess Volatility (sEV) and how is it different from just measuring forecast changes?","acceptedAnswer":{"@type":"Answer","text":"Scaled Excess Volatility (sEV) is a probabilistic forecast volatility metric that only penalizes revisions moving a forecast away from ground truth or overshooting it, unlike naive percentage-change metrics that treat every revision as equally bad. It assigns zero penalty to accuracy-improving revisions, full quantile-loss penalty to deteriorating ones, and a partial penalty for overshoots, distinguishing beneficial revisions from harmful volatility. Teams weighing new volatility metrics for forecasting pipelines can track this research area on daily.dev."}},{"@type":"Question","name":"How much does exponential smoothing ensembling reduce forecast volatility for forking-sequences models without hurting accuracy?","acceptedAnswer":{"@type":"Answer","text":"Applying exponential smoothing at inference with alpha=0.9 reduces scaled Excess Volatility (sEV) by 13.2%, 13.0%, 10.9%, 10.2%, and 11.2% for RNN, LSTM, CNN, Transformer, and StateSpace-based encoders respectively, while degrading forecast accuracy (sCRPS) by less than 0.1%. Lower alpha values cut volatility further but at greater accuracy cost. Engineers tuning forecast ensembling trade-offs can follow developments like this on daily.dev."}},{"@type":"Question","name":"Can forecast ensembling reduce volatility on pretrained time series foundation models like Chronos-2 or TimesFM without retraining?","acceptedAnswer":{"@type":"Answer","text":"Yes, forecast ensembling can be applied zero-shot to pretrained models originally trained with window-sampling, including Chronos-2, Toto 2.0, TimesFM, PatchTST, and N-BEATS, by aggregating their forecast revision outputs. Across the M-series benchmark this achieved a median roughly 10% reduction in forecast volatility with less than 0.1% degradation in accuracy (sCRPS). Practitioners evaluating foundation models for forecasting can keep up with such techniques via daily.dev."}}]} ```