TimeSeries

A modular research toolkit for time series econometrics focused on monetary policy and investment dynamics. It implements multiple approaches to handle mixed frequencies and proxy identification, inspired by Diego Känzig’s course.

Highlights

• Mixed-Frequency VAR (MFVAR) with custom Kalman filter and aggregation constraints
• MIDAS regression to temporally disaggregate quarterly variables to monthly
• Simple interpolation VAR (baseline replication)
• Proxy SVAR identification with bootstrap inference
• Investment subcomponents (equipment vs structures) analysis
• Consistent, modular interfaces and comparison pipelines

Current status: Interpolation VAR, MIDAS, MFVAR, Proxy VAR, and comparison tooling are implemented and used in notebooks and scripts. Local Projections are partial; Stacked VAR is planned.

Install

Requires Python 3.10+.

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
# If needed, common packages:
# pandas numpy matplotlib statsmodels pandas-datareader openpyxl scipy scikit-learn jupyter

Quick start

Use the high-level pipelines to run standard analyses and compare approaches.

from src.pipelines import quick_var_analysis

comp = quick_var_analysis(
	approaches=["interpolation", "midas"],
	shock_measure="gk"
)
comp.plot_comparison("log_inv")

Or full control via VARPipeline:

from src.pipelines import VARPipeline

pipeline = VARPipeline(data_dir="./data")
results = pipeline.estimate_var_irfs(
	variables=["gs1", "logcpi", "logip", "ebp"],
	shock_measure="gk",
	approach="midas",           # or "interpolation"
	midas_method="beta1",       # 'ols', 'almon', 'almon1', 'beta', 'beta1', 'exp'
	midas_lags=6,
	var_lags=12,
	horizon=48,
	n_boot=1000,
	bootstrap_method="mbb",
	proxy_sample_start="1991-01-01",
	proxy_sample_end="2012-06-30"
)
pipeline.plot_irfs()

See API_GUIDE.md for a comprehensive usage guide.

Project structure

TimeSeries/
├── README.md
├── API_GUIDE.md               # Full API and workflow reference
├── pyproject.toml
├── requirements.txt
├── examples.ipynb              # Usage examples
├── src/                        # Modular implementation
│   ├── __init__.py
│   ├── data_loader.py          # DataLoader: shocks, FRED, merge, prep
│   ├── var_models.py           # ProxyVAR, VARWrapper, MixedFrequencyVAR
│   ├── mixed_freq_var.py       # Custom MFVAR with Kalman filter
│   ├── midas.py                # MIDAS regression (weighting schemes)
│   ├── irfs.py                 # IRF estimation (bootstrap, plotting)
│   ├── pipelines.py            # VARPipeline, ComparisonPipeline
│   ├── utils.py                # Plotting and export utilities
│   └── pytorch_var_optimized.py  # Optional GPU acceleration
├── notebooks/                  # Method tutorials
│   ├── ProxyVAR.ipynb          # Proxy SVAR identification
│   ├── MIDAS.ipynb             # MIDAS regression
│   ├── MFVAR_with_aggregation.ipynb  # Mixed-frequency VAR
│   ├── VARMAX.ipynb            # VARMAX state-space
│   ├── LP.ipynb                # Local Projections
│   └── aggregation.ipynb       # Time aggregation
├── data/                       # Datasets
└── tests/                      # Test suite

Core modules at a glance

• Data loading: unified loader for GK data, FRED investment, multiple shock measures; quarterly→monthly merge and prep
• MIDAS: OLS and parametric weighting schemes (beta, beta1, almon, exp), optional grouped lags
• MFVAR: state-space with time-varying design (quarterly average of monthly latent states) and support for multiple quarterly variables
• VAR models: standard VAR wrapper plus Proxy SVAR identification
• IRFs: estimation with moving block bootstrap (recommended), plotting, hump-shape checks
• Pipelines: easy single-spec runs and side-by-side comparisons

Notebooks

Start Jupyter and explore the examples:

source .venv/bin/activate
jupyter notebook examples.ipynb

Method-specific tutorials are in the notebooks/ directory.

Data sources

• FRED: Real Gross Private Domestic Investment (GPDIC1)
• GKData.xlsx: Gertler & Karadi (2015) monthly macro series
• GKMPSurprises.xlsx: monetary policy surprise instruments
• Optional: Compustat for micro extensions

Testing

Run tests (selected examples shown):

pytest -q tests/

Some tests are long-running (bootstrap); start with a subset if needed.

Troubleshooting

• Import errors in scripts/notebooks:

  import sys; sys.path.append('.')

• Bootstrap is slow: reduce n_boot, or use bootstrap_method='iid' for quick checks
• MIDAS convergence: try beta1, fewer lags, or enable midas_group_lags=True

Roadmap

• Complete Local Projections module and add tests
• Implement Stacked VAR approach
• Broaden identification options (sign restrictions, external instruments)
• Add ragged-edge handling and nowcasting utilities
• Package for pip/conda with CLI and docs site (mkdocs or Sphinx)
• Extend to arbitrary frequency mixes via generalized aggregation operators

Acknowledgments

Built on insights from Diego Känzig’s course; includes replication and extensions of several approaches (interpolation, MIDAS, MFVAR, Proxy VAR).