Skip to content

jgroth1/Project_1_Coffee_Price_Predictors

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

39 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Project_1_Coffee_Price_Predictors

Project Goals:

  1. Determine the features governing coffee price.
  2. Assess the importance of each feature.
  3. Forcast the price of commodity coffee.

Project Hypothesis 1:

As weather is a major contributer production affecting both yield and disease, weather would be a major predictor of price.

Project Hypothesis 2:

Weather alone is not able to forcast the price and so it would need to be combined with production and export/import data.

Project Hypothesis 3:

As the weather and production data are highly seasonal, Seasonal data decomposed from past data plus scalling from weather and production forcast could be used as features in forcasting the future price.

alt text

Coffee Facts

  • According to the Global Exchange, there are approximately 25 million farmers in over 50 countries involved in producing coffee

  • Seventy percent of coffee beans are Arabica. Although less popular, Robusta is slightly more bitter and has twice as much caffeine.

  • Coffee is a popular commodity -> huge global demand

  • Coffee commodity is one of the volatile commodities -> weather is major contributing factor.

  • Arabica coffee is the world benchmark for coffee futures contracts that trade on the Intercontinental Exchange (ICE).

  • Arabica mostly cultivated in Brazil: 40% of world’s total supply

  • Major exporters are : Columbia, Peru, India, Uganda, Ethiopia, Mexico, Cote Ivoire.

alt text

alt text

Forcasting Results:

Weather data alone did not provide addiquate forcasting ability so hypothesis 2 was used to forcast the data.

Ridge Regression Vs. Lasso Regression Vs. Lasso with Outliers Removed

Ridge Results:

Training Adj. R^2 = 0.905

Testing Adj. R^2 = -8.7

MSE = 1353.5

MAE = $31.02

Lasso Results:

Training Adj. R^2 = 0.683

Testing Adj. R^2 = 0.371

MSE = 87.78

MAE = $7.165

Deoutliered Lasso:

MAE = $12.84

Coeficients for Lasso and Ridge regression:

alt text

Lasso Regression + ARIMA Results:

The final model was made as a combination of the Lasso regression

alt text

alt text

Mean Absolute Error = $7.165

alt text

alt text

Forcasting Procedure and Results:

alt text

alt text

alt text

About

Description of The coffee market and Arabica commodity price forecasting

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published