https://github.com/eohne/bayesianfactorzoo.jl
https://github.com/eohne/bayesianfactorzoo.jl
Last synced: about 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/eohne/bayesianfactorzoo.jl
- Owner: eohne
- License: gpl-3.0
- Created: 2024-11-18T17:58:41.000Z (7 months ago)
- Default Branch: main
- Last Pushed: 2024-12-22T10:41:05.000Z (5 months ago)
- Last Synced: 2025-02-14T06:35:54.323Z (4 months ago)
- Language: Julia
- Size: 851 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# BayesianFactorZoo.jl
[![][docs-latest-img]][docs-latest-url]
BayesianFactorZoo.jl is a Julia port of the R package 'BayesianFactorZoo' ([CRAN link](https://cran.r-project.org/web/packages/BayesianFactorZoo/index.html)) implementing the econometric methods from the paper:
> Bryzgalova, S., Huang, J., & Julliard, C. (2023). Bayesian solutions for the factor zoo: We just ran two quadrillion models. Journal of Finance, 78(1), 487–557. [DOI: 10.1111/jofi.13197](https://doi.org/10.1111/jofi.13197)
If you use this package, please cite the original paper.
For detailed documentation of the original R implementation, see the [R package documentation (PDF)](https://cran.r-project.org/web/packages/BayesianFactorZoo/BayesianFactorZoo.pdf).
Note that function signatures and exposed functions are equivalent to the R version.
## Installation
For now this is not in the general registry (I will register it at some point). In the meantime you can install it directly from my repository.
```julia
using Pkg
Pkg.add("http://github.com/eohne/BayesianFactorZoo.jl")
```
> **For some examples see the "Tutorial" in the documentation**
> [Link to Tutorial](https://eohne.github.io/BayesianFactorZoo.jl/dev/tutorial/)
## Main Functions
### BayesianFM
```julia
BayesianFM(f::Matrix{Float64}, R::Matrix{Float64}, sim_length::Int)
```
Performs Bayesian Fama-MacBeth regression.
Parameters:
- `f`: t × k matrix of factors (t periods, k factors)
- `R`: t × N matrix of test asset returns
- `sim_length`: Length of MCMC chain
Returns:
- `lambda_ols_path`: MCMC draws of OLS risk premia
- `lambda_gls_path`: MCMC draws of GLS risk premia
- `R2_ols_path`: MCMC draws of OLS R²
- `R2_gls_path`: MCMC draws of GLS R²
### BayesianSDF
```julia
BayesianSDF(f::Matrix{Float64}, R::Matrix{Float64}, sim_length::Int; kwargs...)
```
Performs Bayesian estimation of linear stochastic discount factor (SDF). Supports both flat and normal priors for risk prices.
Parameters:
- `f`: t × k matrix of factors
- `R`: t × N matrix of test asset returns
- `sim_length`: Length of MCMC chain
- `intercept`: Include intercept (default: true)
- `type`: "OLS" or "GLS" (default: "OLS")
- `prior`: "Flat" or "Normal" (default: "Flat")
- `psi0`: Prior hyperparameter (default: 5.0)
- `d`: Prior hyperparameter (default: 0.5)
Returns:
- `lambda_path`: MCMC draws of risk prices
- `R2_path`: MCMC draws of R²
### Model Selection and Testing
#### continuous_ss_sdf
```julia
continuous_ss_sdf(f::Matrix{Float64}, R::Matrix{Float64}, sim_length::Int; kwargs...)
```
Performs SDF model selection using continuous spike-and-slab prior.
Parameters:
- `f`: t × k matrix of factors
- `R`: t × N matrix of test asset returns
- `sim_length`: Length of MCMC chain
- `psi0`: Prior scale (default: 1.0)
- `r`: Spike component (default: 0.001)
- `aw`, `bw`: Beta prior parameters (default: 1.0)
- `type`: "OLS" or "GLS" (default: "OLS")
- `intercept`: Include intercept (default: true)
Returns:
- `gamma_path`: MCMC draws of model inclusion indicators
- `lambda_path`: MCMC draws of risk prices
- `sdf_path`: MCMC draws of SDF values
- `bma_sdf`: Bayesian Model Averaged SDF
####
```julia
dirac_ss_sdf_pvalue(f::Matrix{Float64}, R::Matrix{Float64}, sim_length::Int, lambda0::Vector{Float64})
```
Parameters:
- `f`: t × k matrix of factors
- `R`: t × N matrix of test asset returns
- `sim_length`: Length of MCMC chain
- `lambda0`: Vector of risk prices under null hypothesis
- `psi0`: Prior scale (default: 1.0)
- `max_k`: Maximum model size (optional)
Returns:
- `gamma_path`: MCMC draws of inclusion indicators
- `lambda_path`: MCMC draws of risk prices
- `model_probs`: Posterior model probabilities
## Example Usage
```julia
using BayesianFactorZoo
# Generate example data
t, k, N = 100, 2, 25
f = randn(t, k) # Factor returns (100 periods × 2 factors)
R = randn(t, N) # Test asset returns (100 periods × 25 assets)
# Bayesian Fama-MacBeth regression
bfm_result = BayesianFM(f, R, 10000)
println("Mean OLS risk premia: ", mean(bfm_result.lambda_ols_path, dims=1))
println("Mean OLS R²: ", mean(bfm_result.R2_ols_path))
# Bayesian SDF estimation with normal prior
bsdf_result = BayesianSDF(f, R, 10000,
intercept=true,
type="OLS",
prior="Normal",
psi0=5.0)
# Model selection with continuous spike-and-slab prior
# Set prior Sharpe ratio through psi0
psi_value = psi_to_priorSR(R, f, priorSR=0.1)
css_result = continuous_ss_sdf(f, R, 10000,
psi0=psi_value,
r=0.001)
# Check factor inclusion probabilities
inclusion_probs = mean(css_result.gamma_path, dims=1)
println("Factor inclusion probabilities: ", inclusion_probs)
```
## License
This package is licensed under GPL-3, as required due to being a port of the GPL-3 licensed R package BayesianFactorZoo.
## Related Links
- [Original Paper](https://doi.org/10.1111/jofi.13197)
- [R Package on CRAN](https://cran.r-project.org/web/packages/BayesianFactorZoo/index.html)
## Speed
The length of monte-carlo simulations was set to 100,000 and the example data of the R package was used. I ran each function 10 times and report the median values.
Timings are from a Laptop with a Ryzen 9 6900HS using 8 threads. Multi-threading of the Julia version is indicated by *.
| Method | Julia Time (s) | R Time (s) | Speed Improvement |
|--------|---------------|------------|-------------------|
| Two Pass | 0.0005 | 0.0025 | 5.0x |
| SDF GMM | 0.0011 | 0.0191 | 17.4x |
| Bayesian FM* | 3.0535 | 48.6653 | 15.9x |
| Bayesian SDF* | 3.6099 | 36.8375 | 10.2x |
| Continuous SS SDF | 20.3824 | 50.6916 | 2.5x |
| Continuous SS SDF v2 | 20.9998 | 57.7819 | 2.8x |
| Dirac SS SDF P-value* | 2.9513 | 54.7477 | 18.6x |
[docs-latest-img]: https://img.shields.io/badge/docs-latest-blue.svg
[docs-latest-url]: https://eohne.github.io/BayesianFactorZoo.jl/dev/