Data over time
Business data arrives over time. It has a mixture of trend and noise, order and disorder, pattern and chaos. You want to tease out what the underlying trends are, and make allowance for the noise. You want to make forecasts of what’s most likely to happen, while also accounting for uncertainty.
Our team has applied forecasting techniques to a wide variety of problems across industries, such as forecasting the drone positions, forecasting hemoglobin levels, and forecasting wallet sales.
Forecasting
Time series analysis is a tool for separating the signal from the noise. It helps you to make intelligent forecasts in the face of uncertainty. It helps you answer questions such as the following.
- How much confidence should you have in an apparent trend in your data?
- When things are moving up and down together, how can you tell whether one causes the other or whether both are a result of something else such as a seasonal pattern?
- How does uncertainty increase as you look further ahead?
From conventional to cutting edge
Conventional methods are conventional for a reason. They work well on a variety of problems, are well understood, and easy to implement. So before trying more sophisticated methods, we believe in trying conventional methods first, methods such as LOESS and LOWESS, ARMA and ARIMA.
But sometimes conventional methods are not adequate. Maybe your problem is more structured and could benefit from methods that incorporate your knowledge of that structure, methods that incorporate a dynamical model as well as a statistical model. We might start by building a linear Kalman filter, then if necessary explore nonlinear extensions such as EKF, or work our way up to a particle filter model. The complexity of the model is determined by the amount of exploitable structure in the problem and the forecasting accuracy needed.
We can help
If you’d like help making sense of your data and forecasting what’s likely to happen next, give us a call.
Trusted consultants to some of the world’s leading companies
