This is a simple framework for analysing weather data, with the goal of being able to price simple weather derivatives.
A common type of weather derivative is one based on temperature, defined in terms of monthly Heating-Degree-Day (HDD) index values. The HDD for any particular day is the number of degrees that the temperature on that day was below some threshold (generally 65F in the US). So, for example, if the temperature was 50F, the HDD is 15F. The monthly HDD is the sum of this value over all days in the month.
See here for a more detailed explanation, or
co.abarr.weather.temp.Hdd for the implementation.
There are many possible ways of attempting to predict what the HDD will be over a future time period.
Included here is an implementation of the Alaton Temperature Model, as defined in On Modelling and Pricing Weather Derivatives by Alaton, Djehiche and Stillberger. This can be used to estimate (via Monte Carlo method) the probability distribution of temperatures for a given time period (and therefore the distribution of the HDD index.)
The basic idea of this model is:
- Fit (via linear regression) a simple expected temperature model, based on a sin function with a 365-day period,
- Calculate the residual temperatures by subtracting the expected from observed temperature on each day,
- Fit an Ornstein-Uhlenbeck model to these residuals. This is a mean-reverting model, where the temperature updates randomly each day, but over time reverts to the long-term average (ie the expectation from step 1).
See co.abarr.weather.temp.predict.Alaton.
Also included is a parser for historic bulk downloads from OpenWeather. These are CSVs that include a bunch of different weather metrics for a given location: temperature, rainfall, humidity etc, sampled at half-hourly intervals. Only temperature fields are loaded currently.
See co.abarr.weather.owm.FromCsv.
The project is built using Gradle. To compile the code and run all unit tests:
$ gradle test