Forecasting today: where are we?

There are various types of forecasting: corporate finance, weather, political polling, etc. Some are probably using new AI tools to aid in the prediction business.

Business forecasts

I’ve done a lot of the corporate finance type forecasting where I would send to corporate what the team or location thought would be the financial result for the month, quarter, year or whatever period of time requested by corporate. It could be a forecast for a business line, a forecast for a program, a forecast for a refinery, etc.

And at least in the corporations that I’ve been in, senior management would exhort us to do better because they made decisions based upon our forecasts. We were terrible at it.

I don’t think anybody was good at it.

Now my stance was, especially in the tech industry, the industry changes so fast that it is hard to predict. They shareholders… they can just choose another company to invest in if they didn’t like the forecast because the tech industry was never going to be stable.

Looking back, I would say most of my forecasts have been good in that I would come in under 5% variance. But I had some really amazing forecasts too, like the time I wrote about a merger and a few other times coming within $5 dollars of the forecasted revenue.

Freaking amazing results there.

I also, unfortunately, have encountered situations where I thought the forecasts might have driven the accounting. In other words, there were so much pressure in the company that people started to make rationalizations in their head for doing the accounting they did. I saw a couple of booked entries that didn’t look right.

And in the last tech company, somebody gave accounting one set of figures to use for booking revenues and gave another set of figures to the billing agent for their invoices. The finance person accrued to much revenue, based upon the information received, such that when the project ended and finance and billing were closing up the account, they found that the revenue had to be written down, I think, half a million dollars.

That led to our division to being put on the selling block.

Today, there is supposed to be some form of predictive analytics that is said to be markedly superior to what humans have done. If my history is filled with poor forecasting by financial analysts is common throughout the business industry, then there would be no surprise that predictive analytics would be better.

Not perfect but better than what has been done before.

Unfortunately, I have not worked in companies that are deploying these predictive models to be able to make that assessment. I would love to know what goes into those models.

It sounds like there has been some improvement in forecasting in businesses.

If somebody had a large enough model to incorporate all of the factors going into the state of the financial world, maybe the model would have recognized a pattern amongst credit default swap, collateralized debt obligations, mortgage lending, credit rating, and whatever else signaling something amiss. Maybe recognized an outlier pattern.

Such a predictive tool would probably work only if it had a lot of data and then it would have to be told to find an outlier to be investigated as a potential issue. But no human would be able to sift through all of that data and see a big picture pattern; only a powerful machine such as AI.

But no AI would be able to predict the coronavirus that popped out of China in 2020 and led to a surge in unemployment worldwide. That is a swan event that just pops out of nowhere.

I would think predictive analytics tools would work the same way: with enough historical data, the tool could make good regular business forecasts but fail the coronavirus type events. For the swan events, scenario planning could play a role: do some planning with what-if scenarios of what could hit our business and make plans on how to prepare and react, if necessary.

Weather forecasts

This summer was supposed to be the year of the blockbuster hurricane season. The weather people have said that this season was supposed to be the busiest ever and that Texas was in for a terrifying time.

The early part of the season suggested that the busiest season scenario was playing out: Hurricane Beryl formed somewhere in the Atlantic or Caribbean, ramped up to a destructive Cat 5 – earliest ever in the season – and went on to ravage the Caribbean and strike Houston as a Cat 1, and causing other rain miseries across the US.

And then the oceans quieted down after Ernesto in August. After Ernesto, all went quiet which was shocking because the national weather service had projected 17 to 25 storms to form, with 8 anticipated to be hurricanes and 4 to 7 major hurricanes. (From NPR, 2024 will be the most active hurricane season in the Atlantic : NPR.

The average is typically 14 storms.

Everybody was asking what happened.

So did their forecast derail?

Well, we still have rest of September, October and November for the terrifying season to ramp up and we now have multiple tropical waves sitting in the Atlantic and Caribbean. We have a disturbance that just hit the Gulf of Mexico and has a 90% chance of developing into at least a tropical storm, if not a hurricane. Its path is to move north up the coastline – maybe slip into Mexico, thus weakening it, but not really projected to do so – and then curve to the east, and maybe impact the area from Houston to Louisiana. Right now, Louisiana is the main projected landfall, but southeast Texas is not out of the woods yet.

The projected landfall is Wednesday night, Thursday morning.

These tropical waves may be the start of intense activity so we can’t say the forecast failed. We won’t know until end of November.

But I will say, it appears that regular weather forecasting has improved, and they do seem to be better at forecasting a development and the track. The weather people have gathered a lot of weather data and maybe they have reached the tipping point where the models are starting to give better results.

The models do fail at the rapid intensification because that type of hurricane activity has not happened much in the past. It is feared with the warming world, though, that we will face more rapid intensification (and leaving us less time to prepare if the intensification occurs in the Gulf of Mexico).

Political forecasts

Ever since 2016, where everybody thought the polls suggested Hillary Clinton got it in the bag and was thus hugely surprised at the results.

And we still talk about polls today.

Right now, my impression is that polls have improved with Harris at the top, but they still waffle all over the place. I think right now the polls are too close to say who will be the winner.

And that is a stunner because it is a choice between the authoritarian and the freedom/democracy fighter.

I don’t know if anybody is using AI for polls. But there are two notable forecasting agency that have a record of being spot on in their picking of winners: 538 and Allan Lichtman’s 13 keys.

As of today (9/8/2024), 538 has Harris leading in the polls but very slimly, maybe within the margin of error. This is in the national polls which I don’t think is valid because the Electoral College is based upon the states. But 538 has a very good history of nailing it.

Allan Lichtman apparently have been using his 13 keys to predict who would win the Presidency since 1984 and has only failed once (Bush-Gore 2000). That’s what I’ve heard.

That 2000 election was basically decided by the Supreme Court, whose, according to reports, Republican leaning judges put their thumbs on the scale. That is how I recall what transpired.

The 13 keys appear to predict very well, except under conditions of “shenanigans”. Right now, he is predicting Harris to win but I don’t think he is including the impact of Republicans’ future efforts to challenge (decry) the results and maybe state election deniers not do their job and confirm the votes in time.

The situation is very different today, with players trying to implement rules that make it harder to vote, doing voter purges, maybe some state actions such as not confirming the votes in time, and court challenges going all the way up to the Supreme Court. Those elements could tip the results the other way.

We are in a different era.