Introduction of Nadeem Shahadi and Nassim Taleb:
Nadeem Shahadi, Director of the Ferris Center for Eastern Mediterranean Studies, introduced Nassim Taleb as the keynote speaker. Shahadi highlighted Taleb’s expertise as a derivatives trader, philosopher, and mathematician, emphasizing his influential books translated into 36 languages.

Taleb’s Arrival and Opening Remarks:
Taleb expressed his surprise and honor at being invited to speak before an audience familiar with his work. He expressed skepticism about the overuse of data and data analytics in predicting events.

Questioning the Role of Data in Prediction:
Taleb emphasized the abundance of data and correlations in finance before the 2007 financial crisis, questioning the usefulness of data in making accurate predictions. He shared an anecdote about testifying in Washington against increased spending on data collection.

Taleb’s Approach to Data and Prediction:
Taleb argued for a “via negativa” approach, where irrelevant data is removed to focus on the most significant factors. He contrasted this approach with the tendency to overload predictions with excessive data, which can lead to confusion and increased risk-taking.

Overview of the Lecture:
Taleb introduced the topics he would cover in his lecture, including the problem of skin in the game in economics, the notion of fragility as an alternative to predictive analytics, and the concept of fat tails.

Fat Tails Explained:
Fat tails refer to the higher probability of extreme outcomes in a distribution. This means that while extreme events are less frequent, they have a significant impact. In contrast to common misconceptions, fat tails do not imply more frequent extreme outcomes, but rather their greater consequences.

Identifying Fat Tails:
Fat tails can be identified by observing that a small number of observations or events account for a large portion of the distribution. In financial markets, for example, a small number of companies generate the majority of sales and profits. Similarly, in the pharmaceutical industry, a small number of drugs account for the bulk of sales.

Fat Tails and the Law of Large Numbers:
The law of large numbers, commonly taught in schools, suggests that as the sample size increases, the average of the sample will approach the true population mean. However, this assumption breaks down in the presence of fat tails, where small deviations can have a disproportionate impact on the distribution.

Implications of Fat Tails:
Fat tails have significant implications for various fields, including finance, economics, and risk management. Understanding fat tails is crucial for accurately assessing risks and making informed decisions in these domains. The presence of fat tails challenges traditional statistical methods and assumptions, requiring alternative approaches to account for extreme outcomes.

Fat Tails and Large Deviations:
In certain domains, large deviations from the average are common, resulting in a fat-tailed distribution. In a fat-tailed distribution, the maximum value is more significant than the average. A small number of observations often represent a large portion of the total value.

Super Fat Tails:
Some distributions exhibit super fat tails, where a small number of events account for a majority of the observations. Examples include the concentration of wealth among a small percentage of individuals and the occurrence of major historical conflicts.

Globalization and Fat Tails:
Globalization has contributed to the emergence of fat-tailed distributions. Winner-take-all effects in a globalized world lead to extreme concentrations of wealth and power.

Ecological Diversity and Fat Tails:
In ecology, the diversity of species is proportional to the size of the habitat. Larger habitats support more species, but the diversity per unit area is lower. Fat tails in ecology arise from the loss of diversity due to various factors.

Unpredictability and the Turkey Problem:
Fat tails contribute to unpredictability and the occurrence of sudden, unexpected events. The “turkey problem” illustrates how conventional statistical methods can fail to predict such events.

Limits of Conventional Data Analysis:
Conventional data analysis methods often fail to capture the characteristics of fat-tailed distributions. Sampling the mean may not provide an accurate representation of the mean in a fat-tailed distribution. Big data can be useful for identifying extreme events, but only when used judiciously.

Targeted Data Analysis:
Data analysis should be targeted to specific questions and objectives. Extreme data points can be valuable for certain applications, such as terrorism detection. However, such data may not be suitable for predicting socioeconomic events.

Importance of Engineering Analysis:
Engineering failures should be analyzed based on structural factors rather than superficial observations. This approach helps identify the underlying causes of failures and prevent future incidents.

Fat Tails and Delayed Blow-ups:
Fat tails in distributions mean that extreme events occur more frequently than expected under traditional statistical models. Delayed blow-ups refer to situations where hidden risks accumulate over time and then suddenly manifest in catastrophic events.

The No Skin in the Game Principle:
When individuals or organizations reap the benefits of an action without bearing the potential adverse consequences, they have “no skin in the game.” This can lead to reckless decision-making and an increased likelihood of financial blow-ups. Examples include excessive bonuses for short-term profits, bureaucracies hiding risks to improve job performance, and corporations passing on hidden losses to taxpayers.

Consequences of No Skin in the Game:
Encourages statistical manipulation and the hiding of risks. Leads to a preference for steady profits over volatility, even if the latter is more stable in the long run. Results in situations where those with the highest rewards also have the highest risks, as seen in the collapse of Bear Stearns funds.

Hedge Funds and Skin in the Game:
Hedge fund managers are forced to “eat their own cooking” as they invest their own money alongside their clients’. This alignment of interests incentivizes responsible decision-making and reduces the likelihood of delayed blow-ups.

Agency Problem and Fools of Randomness:
“Crooks of randomness” refers to individuals who exploit the system for personal gain, while “fools of randomness” genuinely believe their own erroneous ideas. In economics, the agency problem is acknowledged, but its implications in the context of fat tails are often overlooked.

Skin in the Game as an Evolutionary Mechanism:
Unlike economic models, nature operates on the principle that individuals who endanger others must also bear the same risk. This concept is exemplified by the rarity of reckless drivers on highways, as they face the potential consequences of their actions.

Filtering of Individuals Who Endanger Others:
Individuals who recklessly endanger others often meet an early demise, leading to their elimination from the population. Entrepreneurs who make mistakes and bellicose leaders who engage in warfare often face similar fates, preventing their proliferation. Suicide bombers, while dangerous, pose a relatively low risk compared to other causes of death.

Skin in the Game in Financial Markets:
Pit traders serve as an example of individuals with skin in the game, as they risk their own capital and are incentivized to manage risk effectively. The inability to measure risk accurately necessitates a focus on mechanisms that filter out individuals who engage in excessive risk-taking.

Fragility as a Concept:
The speaker introduces the concept of fragility as a key aspect of understanding risk. A simple test is proposed to assess fragility, suggesting that the approach to risk assessment needs to be re-evaluated.

Data Events Don’t Necessarily Have Predecessors:
Greenspan claimed that the financial crisis was unprecedented and couldn’t have been predicted. A simple test on an axis of events shows that tail events often occur without predecessors or successors. This means that past data cannot always predict future events, especially extreme events.

Importance of Engineering:
Engineering is a field where there is no room for “bullshit.” Engineers must consider all the possibilities and design systems that can withstand extreme events. Taleb’s paper for the RAND Corporation focused on the importance of engineering in preventing disasters.

Definition of Fragility:
Fragility refers to a system’s sensitivity to disorder. Fragility is measured by its inability to withstand shocks or unexpected events. Fragility is a universal property that can be applied to various systems.

Measuring Disorder:
Disorder can be measured using different metrics depending on the field. Entropy is a measure of disorder in physics. Standard deviation is a measure of disorder in finance. Time can also be considered a measure of disorder as it breaks things down.

Universal Metric of Disorder:
There is a universal metric of disorder that can be used across different disciplines. This metric is independent of the mathematical method used to measure it. Proving the metric in one discipline proves it in all other disciplines.

Properties of Fragile Systems:
Fragile systems are sensitive to shocks and unexpected events. Fragile systems do not like disorder and are easily disrupted by it. Examples of fragile systems include coffee cups and complex financial systems.

Defining Fragility:
Fragility is a concept that describes the negative impact of shocks on a system, resulting in disproportionately large losses compared to the benefits gained. Fragility is characterized by a concave relationship between the intensity of shocks and the resulting harm. A simple example of fragility is a coffee cup, which can be easily broken by a shock, but doesn’t benefit from shocks and has a bigger downside than upside.

Anti-Fragility:
Anti-fragility is the opposite of fragility, where a system benefits from shocks and stressors, becoming more robust and resilient. An example of anti-fragility is a stone, which becomes stronger when subjected to repeated shocks. Anti-fragility is characterized by a convex relationship between the intensity of shocks and the resulting harm.

Fragility and Concavity:
A concave relationship between shocks and harm indicates fragility, meaning that increasing shock intensity leads to disproportionately greater harm. This relationship is often observed in systems with negative feedback loops, where a shock triggers a chain of events that amplifies the initial harm.

Anti-Fragility and Convexity:
A convex relationship between shocks and harm indicates anti-fragility, meaning that increasing shock intensity leads to disproportionately lesser harm or even benefits. This relationship is often observed in systems with positive feedback loops, where a shock triggers a chain of events that mitigates the initial harm or even leads to positive outcomes.

Implications for Policy and Decision-Making:
Understanding the concepts of fragility and anti-fragility can help decision-makers and policymakers design systems and policies that are more resilient and adaptive to shocks. Fragile systems should be avoided or mitigated, while anti-fragile systems should be encouraged and promoted.

Examples of Fragility and Anti-Fragility:
The Soviet Union’s collapse was attributed to its size and centralization, making it fragile to shocks. The comparison of China and Singapore illustrates the significant impact of size on fragility. The pre-unification history of Italy shows that low-grade turmoil can contribute to stability, while large-scale events can lead to instability.

Cyber Security and Fragility:
Cyber security is a major concern, but the industry’s focus on resilience and constant testing can help mitigate fragility. Netflix’s “Chaos Monkey” program, which engineers failures daily, is an example of an anti-fragile approach to system security.

Cyber Security:
Taleb is not as concerned about cyber security as he is about biological risks, such as the spread of disease through Uber taxis. He believes that organizations are spending too much on cyber security and that the cost of controlling these risks may continue to increase. Cyber security can also slow down operations and create indirect costs.

Systems Thinking:
Taleb admits that he does not understand systems thinking and has tried to read about it but found the concepts difficult to grasp. He prefers a mathematical framework for understanding complex systems.

Blind Spots:
Taleb recognizes the importance of identifying blind spots but acknowledges that it is challenging to do so since a known blind spot is no longer a blind spot. He is working on developing methods to improve stress testing and detect accelerations, which could help identify blind spots.

Statistics:
Taleb believes that statistics is often used mechanistically and that statisticians do not always consider tail events. He argues that statistical tools for dealing with tail events are not yet refined and need further development.

GMOs and Silent Risk:
Taleb mentions his work on GMOs and silent risk, which he has made available for free on the web. He criticizes the use of statistics in evaluating GMOs, as statisticians may not fully understand the biological implications of their findings.

Formalizing Risk and Methodologies:
Nassim Taleb emphasizes the importance of formalizing a method and approach to understand risk and uncertainty. He draws parallels between statistics as an application of probability theory and risk as an application of statistics. Taleb highlights the need for a structured framework to identify and assess risks, as verbal descriptions alone may not reveal the true extent of risk.

The Irrelevance of Data in the Body of a Distribution:
When working with tail events, the data in the body of a distribution is irrelevant. Techniques such as peak over threshold and pot method focus on extracting the maximum values from data, rather than analyzing the entire distribution.

Large Deviation Theory and Extreme Value Theory:
Taleb introduces the concept of large deviation theory and extreme value theory, which deal specifically with tail events. These branches of mathematics provide insights into the behavior of extreme values and rare occurrences.

Focusing on Tail Events for Character Evaluation:
Taleb uses the example of O.J. Simpson’s murder trial to illustrate the importance of focusing on tail events when evaluating character. He argues that a single extreme event, such as a murder, is more informative than a series of regular, non-extreme events.

The Fallacy of Big Data and the Value of Anecdotes:
Taleb criticizes the notion of big data, stating that a large volume of data does not necessarily equate to valuable information. He proposes that a single anecdote, if it occurs in a tail event, can be more informative than a million pieces of data from the body of a distribution.

The Problem with Big Data:
Taleb identifies a problem with big data related to the generation of covariance matrices. He argues that generating a covariance matrix with a large number of variables can lead to false correlations and unreliable conclusions.

Curse of Dimensionality:
The curse of dimensionality refers to the phenomenon where the number of spurious correlations increases convexly with the dimensionality of the data. This means that as more variables are added to a dataset, the more likely it is to find spurious correlations, even if there is no real relationship between the variables.

Large D, Small N Problem:
In the context of big data, the problem of large D, small N arises when the number of variables (D) is large relative to the number of data points (N). This problem is particularly problematic because it increases the likelihood of finding spurious correlations.

Spurious Correlations in Big Data:
In big data, the large D, small N problem can lead to the discovery of spurious correlations that appear to be plausible but are not supported by real relationships between the variables. This can be misleading and lead to incorrect conclusions being drawn from the data.

Data Cleaning and Variable Selection:
To mitigate the problem of spurious correlations in big data, it is important to carefully clean the data and select only the most relevant variables for analysis. This can help to reduce the dimensionality of the data and make it less likely to find spurious correlations.

Importance of Inference:
Inference is the process of drawing conclusions about a population based on a sample. In the context of big data, it is important to use appropriate statistical methods to make inferences about the population from the data. This can help to ensure that the conclusions drawn from the data are accurate and reliable.

Challenges with Data Analysis:
Under fat tails, data grows slower than expected, requiring more data for meaningful analysis. The Bonferroni upper bound and the Malesenko-Pasteur bound limit the reliability of results when using multiple variables. Researchers may engage in questionable practices, such as selective reporting of results, to achieve statistical significance, hiding the true extent of their experimentation.

Skin in the Game:
Evaluating the credibility of data analysts: If they work for a living despite their claims of data mastery, it suggests a lack of confidence in their abilities. The most sophisticated data analysts often use their knowledge for short-term trading or inventory management, rather than long-term investments.

Ethical Concerns:
The problem of ethics is central to big data analysis. Researchers may manipulate data or selectively report results to achieve desired outcomes. The FDA, for example, dismisses epidemiological studies conducted solely on computers, highlighting the need for caution in relying on data analysis.

Ineffectiveness of Data Analysis in Socioeconomic Domains:
In socioeconomic domains, data analysis has not consistently led to improved outcomes. The finance industry, despite its access to vast data, often fails to outperform the market collectively, indicating that data alone is insufficient for accurate predictions.

Data Analysis in Different Fields:
Data analysis is valuable in certain fields, such as bird theory, hydrology, and natural sciences, where variables are often linear and predictable. In contrast, data analysis is challenging in non-linear domains, such as economics and finance.

Conclusion:
While data analysis can be a powerful tool, it is important to recognize its limitations and ethical challenges. Researchers should exercise caution in interpreting results, especially in complex and non-linear domains.

Economists’ Reliance on Statistics:
Economists often make assumptions and rely on statistical models, even when these models may not be reliable. This is problematic because it can lead to poor decision-making and financial instability.

The Correlation between Statistics and Bullshit:
The more statistics someone uses to support their argument, the more likely they are to be misleading or incorrect. This is because statistics can be easily manipulated to support a desired conclusion.

The Importance of Robustness:
When making decisions, it is important to consider the robustness of the available data and statistics. If the data is unreliable, it is better to avoid making a decision rather than relying on potentially misleading information.

The Stiglitz Effect:
The Stiglitz effect refers to the tendency of economists to lose their common sense and analytical rigor when they enter academia. This can lead to them making poor decisions and giving bad advice.

The Value of Practical Experience:
Practical experience in risk management and decision-making is more valuable than theoretical knowledge. This is because real-world experience provides a better understanding of the risks involved and how to mitigate them.