Analog Forecasting¶

Why It Matters¶

Complex systems are often too chaotic for mechanistic prediction but not random enough to ignore history. Analogs leverage the insight that similar initial patterns tend to evolve similarly, even when the exact physics is intractable. This provides a reality-based forecast when models are unreliable or unavailable, and generates testable hypotheses about what might happen next.

The Key Move¶

Define what makes a situation "similar" - which features must match (temperature gradient structure, pressure pattern) versus which can vary (exact values, calendar date). Search historical records for the closest analogs based on those criteria. Examine how the analogs evolved - do they cluster toward a common outcome or diverge? Use the analog outcomes as an ensemble forecast. Note: the quality of analog forecast depends entirely on choosing the right similarity metric.

Classic Application¶

Hurricane track forecasting before numerical models. Forecasters found historical hurricanes with similar starting position, intensity, heading, and atmospheric steering pattern. The historical tracks formed a probabilistic forecast. Modern analog forecasting still complements numerical models for rare events (extreme precipitation) where models have limited training and analogs provide grounding in actual physics.

Surprising Application¶

Product launch strategy. Find companies that launched similar products in similar markets with similar resources (adjust for inflation/technology). How did their launches evolve? What worked and what failed? Use this analog ensemble to forecast your own trajectory and identify critical decision points. A SaaS company launching in healthcare can find analogs in adjacent regulated markets rather than guessing from first principles. The analogs reveal timescales, common pitfalls, and resource requirements.

Failure Modes¶

Surface similarity - matching on obvious features that don't drive dynamics (calendar date for weather, company age for startups). Overfitting - requiring such exact matches that you find too few analogs or miss genuinely similar cases that differ in irrelevant details. Small sample size - with only 2-3 analogs, you can't distinguish signal from noise. Non-stationarity - assuming old cases apply to a changed system (climate change making historical weather analogs less relevant). Ignoring unique features - even with good analogs, unprecedented elements matter.

Go Deeper¶

Van den Dool, Empirical Methods in Short-Term Climate Prediction (2007); Zorita & von Storch, "The Analog Method as a Simple Statistical Downscaling Technique: Comparison with More Complicated Methods," Journal of Climate (1999).