Coin Grouping Findings
| ← Coins Grouping Hypothesis | Team 1 Execution |
Hierarchical Clustering Results
Used hierarchical clustering on scaled returns, creating groups.
5 Clusters → 3 Clusters → 4 Clusters
Testing different cluster counts revealed at least 2 distinct clusters:
- Cluster 4: behavior similar to Cluster 3
- Cluster 2: totally different behavior
Cluster Characteristics
| Cluster | Description |
|---|---|
| 1 | Low to middle-low volume |
| 2 | Shitcoins + stablecoin (DFUSDT, DEXEUSDT, HIVEUSDT, PHAUSDT, UXLINKUSDT, USDCUSDT) |
| 4 | Solid coins with big volumes |
Time-Based Correlation Analysis
Explored changes in correlation between coins - better measure of evolving relationships than overall correlation trend.
Method: Monthly correlation changes over 12 months
Result: Very similar behavior across clusters - may not be significant when derived this way.
Cluster Statistics (Normalized)
| Cluster | Close Price Mean | Std | Volume Mean | Std |
|---|---|---|---|---|
| 1 | -0.0 | 0.999 | -0.0 | 0.999 |
| 2 | -0.0 | 0.997 | 0.0 | 0.997 |
| 3 | -0.0 | 0.999 | 0.0 | 0.999 |
| 4 | 0.0 | 0.997 | -0.0 | 0.997 |
Conclusion
Insufficient time to implement proper coin grouping. Would require training multiple models per group with individual optimization - deprioritized in favor of other workflows.