Forecasting (3): What makes good forecast?

Summary

The script discusses the key factors that contribute to effective forecasting, such as understanding the variables influencing the forecast and having sufficient historical data. It highlights that a large sample size of data (like 1,000 data points) typically enhances model performance, though some models, like deep learning, can work with smaller datasets. The script also emphasizes the impact forecasts can have on the variables they predict, citing an example from the shipping industry where forecasted freight rates influenced market behavior and led to a supply surplus, preventing the predicted rate increase. Accurate data and understanding are crucial for reliable forecasting.

Takeaways

• 😀 The accuracy of a forecast depends heavily on understanding the factors that influence the variable being predicted.
• 😀 A good understanding of the influencing variables leads to better forecasting models and results.
• 😀 Having a large amount of historical data is crucial for building effective forecasting models.
• 😀 For neural network models, around 1,000 data points are ideal to produce reliable results.
• 😀 With deep learning (DL) models, smaller sample sizes (such as 20-30 years of data) can still provide useful insights.
• 😀 Daily or weekly data typically require large sample sizes to produce accurate forecasts.
• 😀 The quality of the data used plays a significant role in the performance of forecasting models.
• 😀 Public forecasts can sometimes influence the very factors they are predicting, affecting outcomes.
• 😀 In industries like shipping, public forecast predictions can lead to market behavior changes that invalidate the forecast.
• 😀 Forecasting models work best when the impact of the forecasted information on the market is considered.
• 😀 Forecasting accuracy improves as more data and better-quality data are incorporated into the model.

Q & A

What are the key factors that contribute to a good forecast?

The key factors include understanding the variables that affect the target being forecasted, the availability of historical data, and the ability to use appropriate forecasting models. The better we understand the influencing factors and have quality data, the better the forecast model will perform.

Why is historical data important in forecasting?

Historical data is crucial because it provides the foundation for building accurate forecasting models. The more historical data points available, the better the model can learn patterns and make predictions. For neural network models, at least 1,000 data points are typically recommended.

What is the minimum amount of data needed for neural network models to work effectively?

For neural network models, typically 1,000 data points are considered a good minimum for effective model performance. Below this number, models may struggle to make accurate predictions.

Can forecasting models work with smaller data sets?

Yes, in some cases, forecasting models can work with smaller data sets. For instance, some deep learning models can perform well with as few as 20 to 50 data points, especially when using long-term data like annual data for 20 to 40 years.

How does the frequency of data (daily, weekly) affect the amount of historical data required?

When dealing with daily or weekly data, large samples of historical data are generally required to build an effective model. More frequent data typically means the model needs more data to capture trends and patterns accurately.

What role does data quality play in forecasting?

Data quality is essential in forecasting because the better the data (i.e., more accurate, comprehensive, and relevant), the better the forecast performance. Poor data quality can lead to incorrect predictions and unreliable forecasts.

What is an example of how forecasts can affect the outcome they are predicting?

An example from the shipping industry illustrates how forecasts can affect outcomes. When a forecast predicts an increase in shipping freight rates, ship owners may invest in the market based on that prediction, leading to an oversupply of ships and preventing the forecasted rate increase from happening.

Can public forecasts influence market behavior?

Yes, public forecasts can influence market behavior. If people see a forecast predicting a certain outcome, such as an increase in prices, they may act on that prediction, which in turn can alter the market dynamics and prevent the forecasted event from occurring.

How does the shipping industry demonstrate the impact of forecasts on market outcomes?

In the shipping industry, when forecasts predict an increase in freight rates, ship owners may invest heavily in the market. This increased supply of ships can cause a surplus, which can lead to freight rates not rising as initially predicted.

What are the challenges when forecasting in industries like shipping?

Challenges in industries like shipping include the potential for forecasts to alter market behavior, the need for large historical data sets, and the complexity of accurately predicting variables that may be influenced by external factors or market actions.