What Makes a Peer? Valuation-Anchored Similarity in Private Markets

Battle Count: 60

The paper is primarily relevant to private market investing (PE, VC) rather than public market quantitative trading. However, the supervised similarity learning framework and tree-based proximity concepts have transferable relevance to pairs trading, relative valuation in illiquid markets, and risk factor identification. The methodology could inform quantitative strategies in secondary private market transactions or fund-of-funds analytics.

CatBoost is readily available as an open-source library, and the core methodology (training + leaf-node extraction) is straightforward. However, implementing the tree-importance weighting scheme, constructing the full pairwise similarity matrix for 53,000+ companies (resulting in ~2.8 billion pairs), conformal prediction with heteroskedasticity adjustments, and the k-NN benchmarking pipeline requires significant custom engineering. The data preprocessing for high-cardinality categorical features and multi-label deal types adds complexity.

The paper provides detailed hyperparameters (Table 3), feature descriptions (Table 2), and algorithmic formulations. However, the dataset (~270,000 private companies) is proprietary BlackRock data with no public availability. CatBoost is open-source, and the methodology is well-described, but exact reproduction requires access to the same private market data. No code repository is provided.