16 September 2024

The following question was part of the GJ Open Forecasting website.

***What will be the spot price per barrel for West Texas Intermediate (WTI) crude oil on 30 September 2024? Github***

For the unversed,

West Texas Intermediate (WTI) is a grade or mix of crude oil; the term is also used to refer to the spot price, the futures price, or assessed price for that oil. (Source: Wikipedia)

Crude oil prices are notoriously volatile, influenced by geopolitical events and market supply-demand dynamics. Let’s see how I used a Monte Carlo simulation to predict potential crude oil prices for September 2024 based on historical data.

The historical data is available for download at the following site: FRED. I downloaded the data for one year.

EDA and Basic Statistics

Let's first examine the data we have. Below are the first five rows, showing the date and the corresponding crude oil price. One important observation is that the oil prices are initially recognized as “object" data types in Python. To proceed with analysis, I needed to convert this column into a numeric format to perform calculations. Fortunately, after cleaning the data, I confirmed that there were no missing values, ensuring the dataset was complete and ready for use.

Next we can have a look at the distribution of the oil prices.

Coming to the daily returns, the mean daily returns are -0.000731 and the standard deviation is 0.0189.

Monte Carlo Simulation

To predict future prices, I first calculated the daily returns (i.e., the percentage change in price from one day to the next). Using the historical mean return and the volatility (standard deviation of returns), I generated random daily returns for the entire month of September. I ran 1,000 simulations of possible price paths, each representing a potential scenario for crude oil prices for nearly 17 trading days left in September. (The most recent data available is from September 9th).

The mean predicted price was coming out to be: $69.06 with a median value of $68.80. With the following buckets as possible answers, it is reasonable to split the probabilites into the highlighted buckets: