AI FORECASTING REPORT Series Introduction

522

By Terry Kades, BlueWave Forecasting

If AI is not the magic wand, what is?

Where to invest, what to expect, when to enter and exit? And the real question is “what tools will give the answer” and “how would you know”?

This will be the first of a bimonthly series discussing a variety of current topics relating to AI and scenarios for industry and business, tools that guide investments.

Let’s begin by acknowledging that each forecasting method has strengths and Achilles heels. There simply is no silver bullet when it comes to forecasting. Selecting the best method depends on the amount of and stability of historical data and the predictability of the business environment.

Statistical methods which include AI rely on a minimum amount of historical data which are fed into various statistical modelling and algorithms to generate predictions and forecasts. How much historical data? Well, to be safe and to improve confidence in the forecast, statistical algorithms generally require at about 18 periods of data. If the data is quarterly, then several years of data are necessary.

Think about this…the further back you go in history, the higher the confidence. Does this make sense? Is more recent data not more meaningful?

Variance in the historical data is another major factor that challenges statistical forecasts.

First was COVID, then came China supply chain blockages, then Russia rattling Europe’s energy supply, Middle East wars, then a major drought compromised major shipping route via the Panama Canal. Use of the Suez Canal also impacted by wars, inflation and interest rates, and BRICS…these have all caused many dips, troughs and spikes and disrupted historical data since early 2020.

Can AI autocorrect for all the variances and complexity and guide investment decisions with increased confidence?

To believe the answer to this question, let’s take a look under the hood at the history and workings of AI tools. AI uses statistical models and probably with the most fundamental being Bayesian theory.  Thomas Bayes who died in 1761, was a Presbyterian minister and statistician who developed Bayes’ theorem which still forms the basis of commonly used statistical models. With our modern technology, computing power, feedback loops, and interfaces to other algorithmic models mostly with continual feedback, what is generally called AI can produce astounding results in some areas with ChatGPT as an example that still fascinates. The words and grammar in language give meaning and the ChatGPT algorithms feast on and grow stronger with each enquiry and addition of new and relatable data.

AI can impress with rule-based disciplines such as finance, medicine e.g. reading radiology scans, and some areas of engineering. But AI falls short where human judgement is required. For example, data we analyzed with sophisticated models showed GDP for posts and telecommunications to have high causality with light truck sales. While statistically true, industry experience and human judgement knows this is a false positive. Maybe in the future it will, but right now AI does not have enough data and rules to mimic useful human judgement.

This takes us back to our initial question for investors….Where to invest, what to expect, when to enter and exit? And “which tools will give the answer” and “how would you know”?

The next article will be a deep dive into the world of industry and business scenarios which we see as the golden key that boosts AI to the next level for investors and analysts.


Terry Kades is the CEO and co-founder of BlueWave Forecasting, which provides industry and business scenarios and forecasting. An experienced director and senior management consultant he is expert at identifying the drivers and creating scenarios that guide business plans, enhance analysis of investments and provide accurate forecasts. He is convinced that the true power of AI will flourish when it incorporates properly constructed scenarios that fertilize and boost the intelligence aspect of AI.