Instrumented Principal Component Analysis (IPCA) application to asset pricing based on Kelly, Pruitt and Su (2017 and 2019) to model stock and bond returns.

1. Install requirements from requirements.txt

   pip install -r requirements.txt
   

2. Run main.py
   • At startup, the script requires a WRDS login.
   • The script attempts to download the required datasets automatically from their sources. In case if issues, download manually and provide under data/{dataset}/.
     • fnw: Freyberger, Neuhierl and Weber "Dissecting Characteristics Nonparametrically" (2017), as used by Kelly, Pruitt and Su (2019) "Characteristics are Covariances" Data Source
     • kpbonds: Kelly and Pruitt (2022) "Reconciling TRACE bond returns" Data Source
   • The execution then proceeds in the following order:
     • IPCA Application to Stocks
     • IPCA Application to Bonds
     • Common Factor Structure
   • The script can be executed fully at once or step by step.

   python main.py

3. Test IPCA model results (example usage)
   • Once an IPCA model has been estimated, it can be analyzed as described below.

   model.r2 # R2 on Asset (Rr2) and Managed-Portfolio Level (Rx2)
   
   model.Gamma # Loadings
   
   model.Fac # Factor Returns
   
   model.visualize_factors() # Loadings Visualized
   
   model.visualize_gamma_heatmap() # Factor Returns Visualized
   
   model_bonds.Fac.T.mean() # Factor Expected Returns (monthly)
   
   (model_bonds.Fac.T.mean()/model_bonds.Fac.T.std()) * (12**0.5) # Factor Sharpe Ratios (annualized)

(c) Jonas Knappitsch at Vienna University of Economics and Business (2025)

The results as covered by the underlying Thesis are provided under results/ and illustrated in the following.

model_stocks.visualize_factors()

model_stocks.visualize_gamma_heatmap()

model_bonds.visualize_factors()

model_bonds.visualize_gamma_heatmap()