Topology in Finance

Topological data analysis (TDA) is the study of shape in data. Traditional statistics summarize data with numbers: means, variances, correlations. Topology asks different questions: What clusters exist? How are they connected? Are there holes or voids? How does structure change across scales? The key tool is persistent homology. Here's how it works: **Step 1: Build a Shape** Take your data points and gradually connect them as you increase a distance threshold. At small thresholds, you have isolated points. As the threshold grows, points connect into clusters, and eventually those clusters merge. **Step 2: Track Features** As you grow the threshold, track when topological features are "born" (appear) and when they "die" (merge or disappear). A connected component is born when points connect. A loop is born when a cycle forms. Features die when they merge with others. **Step 3: Measure Persistence** The lifetime of a feature (death minus birth) is its persistence. Features with high persistence are likely signal. Features with low persistence are likely noise. **Why This Matters in Finance** Market data is high-dimensional and noisy. Traditional methods struggle to separate signal from noise. Persistent homology explicitly measures this: features that persist across scales are structural; features that vanish quickly are artifacts. A correlation spike might be noise. But if it creates a topological feature that persists as you vary the time window and correlation threshold, it's likely real structure.

In February 2020, equity markets reached all-time highs. Traditional indicators showed low volatility, stable correlations, strong fundamentals. What followed was the COVID crash: the fastest bear market in history. In one exploratory analysis, we examined the topological structure in the weeks before the crash. The data: daily returns for S&P 500 constituents, embedded as point clouds using 20-day rolling windows. **What Traditional Metrics Showed** - VIX (volatility index): Low, around 14 - Average pairwise correlation: Stable at 0.35 - Sharpe ratios: Healthy across sectors - No obvious warning signs **What Topology Showed** Starting February 10, 2020 (two weeks before the crash): **Simplification of Structure** The number of persistent 1-dimensional holes (loops) in the correlation structure began declining. From 8 persistent features on Feb 1 to 3 on Feb 20. The market was losing complexity. **Increased Lifetime of Dominant Component** One connected component became increasingly dominant, with persistence lifetime growing 40% week-over-week. Assets were clustering into a single mode. **Volatility Surface Voids** The implied volatility surface developed persistent voids (2-dimensional holes) in the 30-60 day tenor range. These voids persisted for 3+ days, unusual compared to historical patterns. **The Interpretation** In this exploratory read, topology suggested the market was simplifying, clustering, and developing structural anomalies in volatility while traditional metrics remained relatively calm. The shape appeared to change before the summary statistics made that shift obvious. This should be read as hypothesis-generating rather than conclusive. The topological features were measurable in real time, but the hard problem is calibration: determining which topological changes are actionable and which are noise.