Forecasting (8): Data setup and naive forecast

Summary

This video introduces forecasting using Excel, focusing on a time series data from 1996 to 1999. The main variable of interest is the freight rate for transporting iron ore from Brazil to China. The presenter demonstrates a dynamic forecast model using the Naive Forecast method, explaining how to train and test the data, and calculates the Mean Absolute Percentage Error (MAPE) to evaluate forecast accuracy. The goal is to improve forecasting models with lower MAPE for both training and test samples, setting the Naive Forecast as a benchmark for comparison with more advanced methods.

Takeaways

• 😀 The video focuses on forecasting using Excel with a time series dataset from 1996 to 1999.
• 😀 The key variable of interest is the freight rate for transporting iron ore from Brazil to China, which is used for forecasting.
• 😀 The dataset contains both supply (number of new ship orders) and demand (China's import data), but the forecast focuses on the freight rate.
• 😀 The training and test sample split follows the 80/20 rule, with 80% of the data used for training and the last six months used for testing.
• 😀 Dynamic forecasting is used, meaning forecasts are based only on past known data without access to future values.
• 😀 The Naive Forecasting Model assumes that the forecast for any period is equal to the value from the previous period.
• 😀 The presenter uses the Naive model to forecast the freight rate and then compares the forecasts with actual data for error evaluation.
• 😀 The Mean Absolute Percentage Error (MAPE) is used to measure forecast accuracy. MAPE compares the forecasted and actual values in percentage terms.
• 😀 The calculated MAPE for the training sample was around 6%, indicating good accuracy for the training data.
• 😀 For the test sample, the MAPE was significantly higher, around 35%, highlighting that the Naive forecast did not perform well for unseen data.
• 😀 The presenter plans to explore more advanced forecasting methods in future videos, aiming to achieve lower MAPE values for both training and test data.

Q & A

What is the primary focus of the video?

The video focuses on forecasting using Excel, specifically with a time series dataset from 1996 to 1999. It demonstrates how to use the naive forecast model and calculate the Mean Absolute Percentage Error (MAPE).

What is the dataset used for forecasting in this example?

The dataset is a time series from 1996 to 1999, containing monthly data. It includes variables such as ship type (Cape size new building orders), China’s iron ore import data, Cape size iron ore rates, and freight rates from Brazil to China.

What is the main variable of interest in the video?

The main variable of interest for forecasting is the Cape size iron ore rates, specifically the freight rate for transporting iron ore from Brazil to China.

How does the video define the two variables—supply and demand?

The video defines new building orders of ships as a supply variable, as it represents how many ships are entering the market. The demand variable is defined by China’s iron ore imports, as China is one of the main importers of iron ore from Brazil.

What does the '80/20 rule' refer to in this forecasting example?

The '80/20 rule' refers to splitting the dataset into 80% for training and 20% for testing. In this case, the training data consists of 39 points, and the test data consists of the last 6 months of the dataset (July to December).

What is a dynamic forecast, and how is it different from a static forecast?

A dynamic forecast is made based on the assumption that no future data is available after the training period. The forecast uses the last known value to predict the next period. In contrast, a static forecast assumes all future values are known, and the forecast is calculated using these complete data points.

How does the naive forecast model work?

In the naive forecast model, the forecast for any given time period is simply the value of the previous time period. For example, the forecast for February would be the actual value from January, and the forecast for March would be the actual value from February.

How is the Mean Absolute Percentage Error (MAPE) calculated in this example?

MAPE is calculated by taking the absolute difference between the actual and forecasted values, dividing it by the actual value, and then averaging the result over all the data points. The percentage error is multiplied by 100 to get the final MAPE value.

What does the video suggest about the MAPE values for the training and test samples?

For the training sample, the MAPE is about 6%, which is considered good. However, for the test sample, the MAPE is about 34-35%, which is much higher and indicates that the naive forecast model is not performing well on unseen data.

What is the goal in the upcoming videos regarding the naive forecast model?

The goal in the upcoming videos is to forecast the same variable using more advanced forecasting methods. The aim is to achieve lower MAPE values for both the training and test samples compared to the naive forecast model, which will serve as a benchmark.