Financial forecasting techniques
Compare statistical models, expert judgment, and machine learning to find a forecasting method that fits your data and goals.
Two approaches to forecasting
Quantitative forecasting uses numerical data and statistical models. Qualitative forecasting draws on expert judgment and market knowledge. Combining the two can help address the weaknesses of each approach.
Quantitative forecasting models
Quantitative models use numerical data. They work particularly well when you have plenty of historical records and expect past trends to continue.
Time series analysis
Time series analysis looks at data collected over time. Moving averages and exponential smoothing are common methods, and many basic market forecasting models start here.
ARIMA models
ARIMA (autoregressive integrated moving average) models relationships between past observations and forecast errors. Differencing can reduce trends; seasonal ARIMA adds seasonal terms. Test accuracy on unseen data.
Monte Carlo simulations
Monte Carlo simulations run thousands of random scenarios to estimate the probability of different outcomes. They help assess risk when outcomes are uncertain.
Qualitative forecasting methods
Qualitative methods help when data is scarce, such as during a global pandemic or after a major technological disruption.
- The Delphi method gathers expert views through questionnaires over two or more rounds.
- Market research uses surveys and consumer feedback to estimate demand for new products or services.
- Scenario planning describes different possible futures so leaders can prepare flexible plans.
Choose a technique
Consider these questions before choosing a method:
- How far ahead do you need to forecast?
- How much historical data do you have, and how reliable is it?
- How precise does the forecast need to be?
- What budget and technical resources can you use?
AI and machine learning
Neural networks and deep learning models analyze nonlinear relationships in financial data. They can also process news articles and social media feeds to update market forecasts in real time.
AI models can be difficult to interpret. People still need to oversee their use and take responsibility for decisions based on their output.
Common forecasting mistakes
Watch for these errors when building or reviewing a forecast:
- Overfitting happens when a model is so closely fitted to old data that it performs poorly on new data.
- Selection bias can arise when you use only data that supports a conclusion you already favor.
- Unusual observations can be real events or errors. Removing all of them can hide risks; investigate before changing the data.
Our home page explains how forecasting fits into business planning and investment decisions.