Back

Can You Forecast a Currency?

Testing whether inflation explains the USD/CAD rate, then forecasting it with ARIMA.

PythonTime-Series ForecastingARIMAStatsmodelsPandasEconometrics
View Code
Year
2025
Context
6-person MSc group project · Business Forecasting module · University College Cork
My role
Data analysis and visualisation: analysed the historical exchange-rate data and built the charts presenting the forecast findings.
Links
Code available on GitHub

The Problem

A famous economic theory, Purchasing Power Parity, says exchange rates should follow inflation differences between countries. Sounds neat. Does it survive contact with real data? We pulled 11 years of monthly US/Canada data from the Federal Reserve and put the theory on trial.

The Approach

  1. 1

    Get clean data

    Monthly USD/CAD exchange rates and consumer price indices for both countries, 2010–2020, from the Federal Reserve's official database. 132 months, zero missing values.

  2. 2

    Test the theory

    Regression of exchange rates against inflation differences. Long-run version: weak (18% explained). Short-run: no statistically significant effect at all. Conclusion: inflation alone doesn't drive this currency pair.

  3. 3

    Prepare the series properly

    Checked whether the data was stable enough to model (it wasn't), and transformed it until it was, the unglamorous step that makes or breaks time-series forecasting.

  4. 4

    Compare six models, forecast with the best

    Scored six ARIMA variants on fit, simplicity, and whether their errors looked random. The winner used the past two months of movement to predict the next, and produced a stable 12-month forecast with a 95% confidence range.

The Evidence

Three series, one picture: inflation is smooth, the exchange rate isn't.
Three series, one picture: inflation is smooth, the exchange rate isn't.
The theory's errors drift and cluster: a model that's missing something.
The theory's errors drift and cluster: a model that's missing something.
Six models scored on evidence; one winner.
Six models scored on evidence; one winner.
12-month forecast: stable, with honest uncertainty bands that widen over time.
12-month forecast: stable, with honest uncertainty bands that widen over time.

The Outcome

0

months of central-bank data analysed

0%

how little of the rate inflation explains

0

forecasting models compared before choosing one

Seen enough?

Let's talk