Time series and actuarial applications project

Financial Time Series and Actuarial Modeling

ARIMA, GARCH, ARIMAX, VAR, and causal analysis across market and macro data

This project shows quantitative depth beyond standard forecasting tutorials. It combines market data, macro series, model diagnostics, causal tests, and forecast comparison to separate what is predictable from what is not.

2,870

Stock observations

23%

Annualized vol

0.0003

Granger p-value

Problem

Time series work is often reduced to automatic forecasting, but finance and macro data require diagnostics, stationarity checks, volatility modeling, and careful interpretation of predictive value.

Approach

Analyzed 2,870 daily stock observations and long-horizon quarterly macro series, tested stationarity, compared ARIMA variants, modeled volatility with GARCH under normal and Student-t innovations, ran Granger causality tests, estimated ARIMAX and VAR models, and evaluated out-of-sample RMSE.

Results

The work showed that price levels behaved like near-random walks, while volatility remained modelable with a GARCH(1,1) Student-t specification. It also identified construction output as a leading macro signal for GDP with a Granger p-value of 0.0003.

What is in the repository

Separated predictive signal in volatility from weak signal in price levels.

Used diagnostics and out-of-sample validation instead of trusting automated model selection blindly.

Combined finance and macroeconomics in a single coherent workflow.

Showed the limits of over-parameterized multivariate models on small samples.

Role and scope

Econometric analysis, forecasting, diagnostics, and quantitative interpretation

Project context

Time series and actuarial applications project

Main stack

Rforecastrugarchvarstseriesggplot2