Issues contributed:
The concept behind our very successful forecasting record explained.
I will present scenarios, and their probabilities, for two major geopolitical burning points:
Israeli-Iran conflict
The Russo-Ukrainian (NATO) war.
We cannot escape of assuming (relative low) probabilities to an “Apocalypse scenario” in each of the two conflicts.
This entry is going to be a bit different from my previous pieces, as I will first explain the principle of scenario forecasting after which I will provide probabilities to some of my scenarios as well as to those provided by GnS Economics. I got the idea for this post from Julius Lehtinen in X. Much obliged.
I hope this entry clarifies at least some of the misconceptions many people currently have concerning scenario forecasting and probabilities. The main point is that economic and geopolitical scenario forecasting relies on different versions of the complexity theory, including chaos and game theory. Yet, no mathematics-based model of handling the extreme uncertainty and chaos in scenario forecasting (of economic and geopolitical events) currently exists. It remains to be seen, whether such models can be development with implementation of AI in the future. I am optimistic. I’ll start with a short introduction to the concept of probability.
Introduction to probability
In statistical models, and in forecasting, everything essentially depends on probabilities. In the heart of all forecasting is the probability distribution of a random phenomenon, like a financial crisis or a geopolitical conflict. If we consider a classical random phenomenon (or variable), a coin toss, there are two possible outcomes: heads or tails. Thus, the probability of a head (or tail) is 0.5. However, it is possible that the average of, let's say, 10000 coin tosses keeps away from the 0.5 for a very long time (through great number of tosses). This forms the probability distribution, whose mean, the expected long-term value is 0.5. The very idea of probability is empirical. We can statistically anticipate, e.g., future economic events only by observing many real life observations forming a probability distribution that can be used to assess the likelihood of those events.
The mean of the probability distribution describes the long-run average outcome. It's thus a very basic forecast on to where some random phenomenon is likely to end up to, after sufficient number of repetitions (like a coin toss). This leads to the law of large numbers, which states that if one draws observations from any population that has a mean, the average of the sample will approach the true mean value of the population. Standard deviation describes the standardized variation of the sample around the mean of the population. These give the central limit theorem, which states that with large enough sample from a population with finite standard deviation, the sampling distribution will follow a normal distribution. This implies that the probability distribution of a random variable, that is, its observations tend to center around the mean with no clear bias to left or right. This is the most frequently used distributional assumption as its simplicity and "well-behavedness" (meaning that its stable) makes building statistical models around it relatively easy.
Yet, assuming a normal distribution in economic, financial and political forecasting carries massive risks. For example, the assumption of normal distribution of observations was behind the failure of 'Value at Risk' (VaR) models, which were used to analyze the riskiness of collateralized debt obligations, or CDO:s. This failed assumption almost brought the global financial system down in 2008.
The problem of assuming normally distributed economic and financial phenomenons is that many of them do not actually follow such a well-behaved distribution, but the contrary. While returns of stocks, for example, may appear to be following a normal distribution concentrated around some positive value during economic expansions, they tend fall very far from it during recessions and especially during crises. Crises tend to cause a Minsky Moment, where asset returns suddenly experience a market-wide "jump" to negative creating massive losses to investors.
When we observe, e.g., returns of financial assets over a period including many crises assuming they are normally distributed, the distribution tends to become described as having fat tails. This implies that the distribution has too many observations on the values far (several standard deviations away) from the mean, which makes the tails of the distribution “heavy". When the tails of the probability distribution are fat, the statistical inference based on it, will be biased, because the assumed probabilities are false.
It's thus obvious that assuming “normality” is not a feasible, when analyzing and forecasting major economic, financial or geopolitical events. With scenario forecasting, the process of reaching reliable probabilities becomes even more tricky.
The Narrative
About two years ago, I explained our succesful forecasting record (noting that no one is correct 100% of the time) in the Epoch Times:
How have we accomplished this, very successful forecasting record?
We use the combination of narrative and modeling to construct scenarios of possible future developments and to assign probabilities on them.
The narrative needs to uncover the imbalances, risks, and other vulnerabilities lurking in the economy and the financial markets. It is used to explain the process (to tell the “story”) of how an economic crisis or any major economic development comes to be. The narrative is crucial because the economy is a ‘living organism’, meaning that it evolves constantly, and without the narrative, which tracks its development, it’s basically impossible to reliably forecast the development of the economy.
Although the narrative is effectively just a story line on future developments, it has to be based on events that have already occurred. In this sense it resembles diagnostics in medicine. To build a realistic narrative, a diagnosis, on the economy, we need to understand why it is where it is now, or its ‘illness’. That is, we need to obtain deep enough knowledge on the economy ("body") and the financial markets ("circulatory system"). This is why, like diagnostics, the construction of the narrative needs to proceed in a step-wise manner. How has this economic expansion come to be? Where are the vulnerabilities that threaten it? What can trigger them? If they are triggered, what happens next? Which factors will amplify and which will contain the crisis? What will be the likely response of the authorities? How successful can they be in curtailing the crisis? What is their "endgame"?
For example, when the coronavirus hit the world economy in February/March 2020 (see our warning at the end of January), a forecaster should have asked, at least, four questions: how big is the bubble in the capital markets, what will (can) the central banks do to keep it from imploding, what can China do to support the world economy and what governments will do?
The correct answers would have been: large, they are likely to do everything possible, China is likely to provide heavy support, and "a lot". Any answer even close to these would have led on the correct path of thinking, which basically indicated that something horrible was happening, but governments and central bankers will do almost anything to stop the shock to turn into a full-blown crisis. This is what we assumed (see this and this), while we erred on how effective authorities would be in postponing the banking crisis.
Scenarios
From the very beginning, we have used three scenarios: the most likely, pessimistic and optimistic in forecasting the developments in the economy. While there are no limits for the number of scenarios, the most efficient way is, first, try to find the lower and upper bounds (best and worst scenarios) and then the most likely scenario. This is because it ‘boxes your view in’. By finding the limits of your reality-based-imagination (in good and bad), a forecaster comes to terms with what is truly possible, i.e., what can come to be in the worst and best cases. After that you can start to "oscillate" between those far-ends to reach the most likely scenario.
Moreover, building the worst-case scenario is relatively easy. Just take the risks (vulnerabilities) you have discovered thus far and imagine that most of them will come to 'fruition' (it tends to be very unlikely that all of them could occur). Then write the narrative based on that "horror story". The optimistic or the best scenario is also relatively easy to construct. Just think that most of the risks will not materialize or that they can be curtailed effectively and write the narrative, or a "fairy tale", based on that. With the most likely scenario is where it gets tricky.
The pillars of any economic or geopolitical forecasting are build on the foundation of extensive and critical analysis, which validity needs to be checked constantly. In principle, every single important piece of information needs to be taken into account. Forecaster need to ask him-/herself concerning every new piece of information, does this alter the narrative and if it does, how? It is naturally demanding to the point of impossibility to include all relevant data, but it should still be tried nonetheless. This practical impossibility (of including “all” relevant information) gives rise to the most likely scenario. Most importantly, the most likely scenario needs to evolve constantly based on the choices of authorities, politicians and investors.
In narrative-based forecasting, attaching probabilities to different scenarios is naturally quite far from an ‘exact science’. There usually is no data to corroborate or debunk all the points in your scenarios. Thus, in scenario forecasting, probabilities should reflect your thinking, whether data-based or not, on the current likelihoods. The most important thing is to be able to change the probabilities or even the scenarios if situation changes drastically. Forecasting is a continuous (“chaotic”) process, which needs to evolve with the world and the new information uncovered.
Current scenarios and their probabilities
Keep reading with a 7-day free trial
Subscribe to Tuomas Malinen’s Forecasting Newsletter to keep reading this post and get 7 days of free access to the full post archives.