Superforecasting: The Art and Science of Prediction

I try to read every non-technical book involving real-world occurrences of probability -- the most common contexts being sports, finance and everyday life -- but this book, on a less common topic, is perhaps the most fascinating I have ever read. The topic is forecasting of geopolitical events, over the next few months or years. How good are experts, and is accurate forecasting teachable? To me it's self-evident that for uncertain future events, forecasts should be in terms of probabilities, not Yes/No guesses, but there is great resistance amongst self-described experts (TV or journalistic pundits, in particular) to doing so -- "use a number and you risk being unfairly blamed; stick with phrases as fluffy as a puff of smoke and you are safe". Tetlock's career has centered on study of accuracy of expert forecasts, and his 2007 book Expert Political Judgment: How Good Is It? How Can We Know? gives a more academic account (bottom line: not very accurate). Superforecasting is a non-technical account of how to become a good forecaster, based in particular on study of the most successful participants in a large 2011-2015 IARPA-sponsored "prediction tournament" open to the public (I participated; our team 4p63 won its league in the final year). There is of course no crystal ball, but there are many useful "ways of thinking" to embrace, and many psychological pitfalls to avoid. And important to learn from your mistakes via post-mortem evaluations. The book is very readable, with lots of stories and examples of experts and amateurs being right or wrong to illustrate good and bad cognitive styles.

This book tells a great story about how a group of unpaid volunteers were able to defeat the most credentialed experts in a forecasting tournament and it goes through the techniques they used to do it. It's a harrowing underdog story. The author does a good job on showing how to predict the future when it comes to financial and socio-political forecasts but he doesn't go far enough in explaining how we could use these techniques in our daily life when it comes to everyday things like whether to save or spend money, and how much, where to go to school, what career to stay in, whether a relationship will last, how long a given business will stay afloat. After all, we make these big decisions based on future forecasts! The author does state that in the beginning of the book that we make forecasts all the time in our lives but I'm not sure to what degree we're able to consciously apply forecasting principles to every-day life situations. He could've given more practical examples if that were the case. He does say "Just as you can't learn to ride a bicycle by reading a physics textbook, you can't become a superforecaster by reading training manuals. Learning requires doing, with good feedback that leaves no ambiguity about whether you are succeeding. " So going off that you can't just expect to automatically become a good forecaster by reading this book. You have to getting out make a lot of forecasts, get feedback, and revise the way you do things accordingly. The problem is I'm not sure how many people reading this book would be motivated to go out of their way to do this. Still I don't want to detract you from reading this book because it truly was a good read. Just reading about the way these superforecasters would think and go about things should inspire us to do the same. They didn't see their views as "treasures to be guarded but as hypotheses to be tested." They were able to look at multiple perspectives and handle the cognitive dissonance (most ideologically driven people could not bear to do like-wise.) They would seek "active open-mindedness" which means they would go out of their way to have other falsify their views so they can sharpen their perspective. They would tap into the "Wisdom of the Crowds" by getting in lengthy internet discussions with other forecasters where they would "disagree without being disagreeable". They had the "growth mindset" which means they treat every failure not as a blow to their ego but as a learning opportunity as they would have lengthy postmortems on their failed predictions. They had the intellectual humility to recognize that reality is complex, but the confidence in their abilities to execute their task in a determined way.....And so on.... So regardless of whether or not you are able to successful apply these principles to your everyday life, this is still an interesting story and we could use the way these superforecasters think as a model to how we should approach our beliefs about the outside world.

I really enjoyed this book a few years ago, and I have come back to offer a review based on my notes at the time and how the insights have settled for me over time. I took away many key concepts for successfully forecasting uncertain events and also some areas I noted for further exploration. Many of the following notes are structured from the authors' insight into the demonstrated practices of repeatedly successful forecasters. The book mentions repeatedly the importance of measurement for assessment and revising forecasts and programs. Many people simply don't create any metrics of anything when they make unverifiable and chronologically ambiguous declarations. The book emphasizes the importance of receiving this feedback on predictions that measurement allows, as there is a studied gap between confidence and skill in judgment. We have a tendency to be uninterested in accumulating counterfactuals, but we must know when we fail to learn from it. If forecasts are either not made or not quantified and ambiguous, we can't receive clear feedback, so the thought process that led to the forecasts can't be improved upon. Feedback, however, allows for the psychological trap of hindsight bias. This is that when we know the outcome, that knowledge of the outcome skews our perception of what we thought at the time of the prediction and before we knew the outcome. The main qualities for successful forecasting are being open-minded, careful, and undertaking self-critical thinking with focus, which is not effortless. Commitment to self-improvement is the strongest predictor of long-term performance in measured forecasting. This can basically be considered as equivalent to the popular concept of grit. Studies show that individuals with fixed mindsets do not pay attention to new information that could improve their future predictions. Similarly, forecasts tend to improve when more probabilistic thinking is embraced rather than fatalistic thinking in regards to the perspective that certain events are inevitable. A few interesting findings that the authors expand upon in more detail in the book: experience is important to have the tacit knowledge essential to the practice of forecasting, and that grit, or perseverance, towards making great forecasts is three times as important as intelligence. Practices to undertake when forecasting are to create a breakdown of components to the question that you can distinguish and scrutinize your assumptions; develop backwards thinking as answering the questions of what you would need to know to answer the question, and then making appropriate numerical estimations for those questions; practice developing an outside view, which is starting with an anchored view from past experience of others, at first downplaying the problem's uniqueness; explore other potential views regarding the question; and express all aspects and perspectives into a single number that can be manipulated and updated. Psychological traps to be aware of discussed in the book include confirmation bias, which is a willingness to seek out information that confirms your hypothesis and not seek out information that may contradict it, which is the opposite of discovering counterfactuals; belief perseverance, also known as cognitive dissonance, in which individuals can be incapable of updating their belief in the face of new evidence by rationalization in order to not have their belief upset; scope insensitivity, which is not properly factoring in an important aspect of applicability of scope, such as timeframe, properly into the forecast; and thought type replacement, which is replacing a hard question in analysis with a similar question that's not equivalent but which is much easier to answer. Researched qualities to strive for as a forecaster: cautious, humble, nondeterministic, actively open-minded, reflective, numerate, pragmatic, analytical, probabilistic, belief updaters, intuitive psychologists, growth mindset. The authors then delve into a bit of another practical perspective on forecasting, which involves teams. Psychological traps for teams include the known phenomenon known as groupthink, which is that small cohesive groups tend to unconsciously develop shared illusions and norms that are often biased in favor of the group, which interfere with critical thinking regarding objective reality. There is also a tendency for members of the group to leave the hard work of critical thinking to others on the team instead of sharing this work optimally, which when combined with groupthink, leads the group towards tending to feel a sense of completion upon reaching a level of agreement. One idea to keep in mind for management of a group is that the group's collective thinking can be described as a product of the communication of the group itself and not the sum of the thinking of the individual members of a group. There are some common perceived problems with forecasting, which receive attention in the book: the wrong side of maybe fallacy, which is the thinking that a forecast was bad because the forecast was greater than 50% but the event didn't occur, which can lead to forecasters not willing to be vulnerable with their forecasts; publishing forecasts for all to see, where research shows that public posting of forecasts, with one's name associated with the forecast, creates more open-mindedness and increased performance; and the fallacy that because many factors are unquantifiable due their real complexity, the use of numbers in forecasting is therefore not useful. Some concepts that I took note of for further research from the book were: Bayesian-based application for belief updating, which is basically a mathematical way of comparing how powerful your past belief was relative to some specific new information, chaos theory, game theory, Monte Carlo methods, and systematic intake of news media. These are concepts that I was particularly interested in from the book based on my own interests and that I have continued to explore. This book was very valuable for cohesively bringing together the above concepts in the context of a compelling story, based on the DARPA research project which was compellingly won by the author's team as a product of the research that led to this groundbreaking book.

This is an amazing book. I had read Thinking Fast and Slow before this. I thought these books pose juxtaposing ideas. Great insights into how human beings make decisions and how to consciously employ system 2.

O livro de Tetlock foi escrito em parceria com um jornalista (Gardner), que, nas palavras de Tetlock, tornou a escrita fluida e a exposição de conceitos complexos, didática e objetiva. O livro percorre várias correntes do pensamento contemporâneo, como a análise de diversas heurísticas, segundo a concepção de Amos Tversky e Daniel Kahneman (tendo o Autor dialogado ao longo da elaboração do livro com este último, em interação definidas como "socráticas", em suas própria palavras). Para não tornar esta resenha maçante, vou me deter em outros dois aspectos apenas. O primeiro é a ideia, inicialmente formulada por Fermi, de decompor problemas complexos em elementos mais simples, permitindo, com isso, estimar fenômenos com grande precisão. Uma versão mais moderna do mesmo método é detalhada por Lawrence Weinstein (em seus artigos e nos livros Guesstimation e Guesstimation 2.0), que não são citados, por razões que desconheço. O segundo é o debate com Nassim Taleb, com quem o Autor inclusive escreveu um artigo em colaboração. Longe da retórica inflamada de Taleb, ele reconhece que fenômenos de previsibilidade nula e a ocorrência de fatos inteiramente inesperados, são parte da vida e do mundo, e têm consequências matemáticas e estatísticas corretamente definidas por Taleb (distribuições com caudas sobremaneira longas ou pesadas), mas que isso não invalida a capacidade de previsão em um mundo onde tais fenômenos, de fato, podem ocorrer (os famosos "cisnes negros"), mas não constituem uma ameaça a qualquer previsibilidade, como argumentado por Taleb. Os conceitos extremos de Taleb tornariam o mundo, de tal modo, inapreensível, a ponto de flutuarmos na pura estocasticidade. Tendo a me alinhar com Tetlock, pois, a despeito da validade das críticas de Taleb, meu próprio trabalho cotidiano é identificar padrões e analisá-los. Obviamente, em um mundo carente de qualquer padrão discernível, tal exercício é inteiramente inútil.

A brief explanation of how the 'good judgement project' established a fairly accurate way of forecasting. It runs you through sequentially so you understand the key techniques while also explaining theory and reasoning behind it. I would say it's worth read 'thinking slow and fast' as some of the key terms and ideas in it are referenced In this.

Aunque el libro se enfoca en hacer forecasts, en realidad se trata de cómo tomar decisiones, hacia donde mirar y qué tomar en cuenta. Es algo tedioso y técnico en ciertos puntos, pero termina siendo un librazo con mucha investigación detras.

Sehr gründliche Recherche, Einbeziehung der wichtigsten Theorien, plausible Schlussfolgerungen. Absolut empfehlenswert, insbesondere für Auswahl und Bewertung von Managern und Politikern