When building a forecast, we tend to spend a lot of time thinking about what's 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴.

What we should really do, is think about what 𝘀𝘁𝗮𝘆𝘀 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲.

High-school physics example: let's say we need to predict the distance a rocket in space travels for the next 4 hours. We have information about its location for the last 8 hours.

𝗗𝗶𝘀𝘁𝗮𝗻𝗰𝗲: 1🚀 3🚀  6🚀   10🚀    15🚀     21🚀      28🚀       36🚀

What do we do?

Maybe we ignore the oldest data, it's least relevant, so we could just look back at the last couple of hours and try to extrapolate. Take the average speed for the last 2 hours and assume that continues? (That's what businesses typically do!).

Or, we could work out its speed for each of the last 8 hours:

𝗦𝗽𝗲𝗲𝗱: 1🚀 2🚀 3🚀 4🚀 5🚀 6🚀 7🚀 8🚀

That's a bit more interesting - suddenly there's a pattern that our brains can easily handle, and now we can see what 𝘀𝘁𝗮𝘆𝘀 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲.

𝗔𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗶𝗼𝗻 = change in speed / time = 1🚀

So now our forecast is simple. In the absence of other information, we have no reason to think that acceleration will change, so the only forecast input we need for our model is 𝗔𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗶𝗼𝗻 = 𝟭, and maths takes it from there. No complex logic required.

(Yes, of course, the rocket might run out of fuel, but you'll just to model that in the next iteration 😜)

In summary:

▶️ Look for historical data for the thing you're forecasting

▶️ Don't be afraid to look a long way back - some drivers stay the same for a very long time (decades)

▶️ Analyse the data and visualise it

▶️ Look for things that stay the same

▶️ Ask if there's any reason that it won't stay the same in future

✅ Forecast away!