Library / Behavioural Finance
Date of review: October 2022
Book author: Daniel Kahneman, Olivier Sibony & Cass Sunstein
Вook published: 2021
Book author: Daniel Kahneman, Olivier Sibony & Cass Sunstein
Вook published: 2021
Noise: A Flaw in Human Judgment by Daniel Kahneman, Olivier Sibony & Cass Sunstein (2021)
This is the second book by the world's best expert on human judgement, author of the previous bestseller Thinking, Fast and Slow and a Nobel Prize winner, Daniel Kahneman. Errors in human judgement come from biases and noise. The book discusses the sources of noise, its types and has many actionable points to improve your judgement.
I finished reading the book in the early summer of 2022, but it took me more than three months to finally write a review on it. The immediate conclusion you draw from the book is that human judgement has many more errors than we think. One reason is human biases (the subject of Kahnemann's first book - Thinking, Fast and Slow). They lead to a relatively consistent error of over- or under-estimation (we tend to sell our profitable positions too early or hold on to losers for too long, we also tend to overestimate the upside, underestimate the downside, and so on). The difference between error and actual results that biases cannot explain is attributed to noise. Noise leads to more scattered outcomes without a specific pattern. It is harder to detect and, consequently, harder to eradicate in your thinking process.
This first conclusion could have been a good summary. However, it misses so many critical insights, especially on how to improve your decision-making, that I spent more time digesting and reviewing the book than actually reading it. I have broken down this review into three main sections: examples of noise in real life, the theoretical foundation of noise and, perhaps, the most relevant for the readers of this blog - actionable points on minimising noise in your decision-making.
This first conclusion could have been a good summary. However, it misses so many critical insights, especially on how to improve your decision-making, that I spent more time digesting and reviewing the book than actually reading it. I have broken down this review into three main sections: examples of noise in real life, the theoretical foundation of noise and, perhaps, the most relevant for the readers of this blog - actionable points on minimising noise in your decision-making.
Part I - How much is noise in real life?
Here are some of the real-life examples of noise discussed in the book:
When noise is fine
Interestingly, noise is sometimes beneficial. Here are some of the situations:
- When wine experts at a major US wine competition tasted the same wines twice, they scored only 18% of the wines identically (usually, the worst ones).
- It is common to obtain significantly different diagnoses from the same physicians when they are presented twice with the same case.
- Kahneman and his colleagues conducted a noise audit at several insurance firms, and their results were staggering. They asked two randomly selected qualified underwriters to estimate a potential loss in a specific case. The audit found that the median difference in underwriting was 55% of the average of the two estimates. For example, when one underwriter sets a premium at $145,000, the other does not set it at $160,000 - but instead quotes $255,000. Even more alarming was that all executives initially underestimated the magnitude of the noise as they were unaware of the vast difference in opinions among experts.
- The book also shares the results of another experiment at a large asset management firm. Senior managers asked "forty-two experienced investors in the firm to estimate the fair value of a stock (the price at which the investors would be indifferent to buying or selling). The investors based their analysis on a one-page description of the business; the data included simplified profit and loss, balance sheet, cash flow statements for the past three years and projections for the next two. Median noise measured the same way as in the insurance company, was 41%."
- Judges have been found more likely to grant parole at the beginning of the day or after a food break than immediately before such a break. If judges are hungry, they are tougher.
- "A study of thousands of juvenile court decisions found that when the local football team loses a game on the weekend, the judges make harsher decisions on Monday (and, to a lesser extent, for the rest of the week)."
- "A study of six million decisions made by judges in France over twelve years found that defendants are given more leniency on their birthday."
- Two researchers, Edward Vul and Harold Pashler had the idea of asking people to answer a question not once but twice. The subjects were not told that they would have to guess again. Vul and Pashler's hypothesis was that the average of the two answers would be more accurate than either of the answers on its own. The data proved them correct. The first guess was closer to the truth than the second, but the best estimate came from averaging the two guesses.
When noise is fine
Interestingly, noise is sometimes beneficial. Here are some of the situations:
- Fashion and art: different films, books, clothes. As Kahneman writes: "No one would want to live in a world in which everyone has the same likes and dislikes."
- Stock market: while this may sound strange, without noise in judgement, all market participants would have had the same views, which would limit trading and liquidity and eliminate opportunities for buying mispriced securities.
- Scientific research where competition is welcome (e.g. finding the best vaccine). When different teams are looking at different angles would lead to a faster breakthrough than if all groups followed the same approach.
Part II - Sources of Noise
Bias or Noise?
Shooting competitions can help understand the difference between Bias and Noise.
Shooting contest experiment
Shooting competitions can help understand the difference between Bias and Noise.
Shooting contest experiment
Source: Kahneman, Daniel; Sibony, Olivier; Sunstein, Cass R.. Noise: The Flaw in Human Judgement
Team A shows near-ideal results. Team B is biased because its shots are systematically off target in a specific segment. The results of Team C are noisy because the shots are widely scattered. Team D is both biased and noisy. Like Team B, its shots are systematically off target; like Team C, its shots are widely dispersed.
A more formal definition of noise is an "undesirable variability in judgments of the same problem."
Interestingly enough, we can make predictions more easily when we face someone's bias than when we deal with noise. For example, before the next member of Team B makes a shot, we can make a bet on the shot landing in the same area as the first five. In the case of Team C (noisy), we will struggle to explain the pattern of shots and will not be able to predict where the next shot will land confidently.
Another interesting point is that you can always tell noise even if you do not know the correct answer, but you will not be able to identify bias unless you know the right answer. To illustrate the point, Kahneman shows how the back of the targets would look for the four teams. Without knowing where the bull's eye is, you would be unable to tell which team shot better - A or B? While it is evident that both Team C and D had scatter shots and thus suffered from noise.
The great thing is that we can improve our judgements even when we cannot verify whether they are right. To do this, we just have to reduce the noise.
The great thing is that we can improve our judgements even when we cannot verify whether they are right. To do this, we just have to reduce the noise.
How the back of the target looks for different teams
Team A shows near-ideal results. Team B is biased because its shots are systematically off target in a specific segment. The results of Team C are noisy because the shots are widely scattered. Team D is both biased and noisy. Like Team B, its shots are systematically off target; like Team C, its shots are widely dispersed.
A more formal definition of noise is an "undesirable variability in judgments of the same problem."
Interestingly enough, we can make predictions more easily when we face someone's bias than when we deal with noise. For example, before the next member of Team B makes a shot, we can make a bet on the shot landing in the same area as the first five. In the case of Team C (noisy), we will struggle to explain the pattern of shots and will not be able to predict where the next shot will land confidently.
Another interesting point is that you can always tell noise even if you do not know the correct answer, but you will not be able to identify bias unless you know the right answer. To illustrate the point, Kahneman shows how the back of the targets would look for the four teams. Without knowing where the bull's eye is, you would be unable to tell which team shot better - A or B? While it is evident that both Team C and D had scatter shots and thus suffered from noise.
The great thing is that we can improve our judgements even when we cannot verify whether they are right. To do this, we just have to reduce the noise.
The great thing is that we can improve our judgements even when we cannot verify whether they are right. To do this, we just have to reduce the noise.
How the back of the target looks for different teams
Source: Kahneman, Daniel; Sibony, Olivier; Sunstein, Cass R.. Noise: The Flaw in Human Judgement
“
When film executives estimate the market for a movie, we can study the variability of their answers without knowing how much the film eventually made or even if it was produced at all. We don't need to know who is right to measure how much the judgments of the same case vary. All we have to do to measure noise is look at the back of the target.
Sources of noise
1). Cognitive abilities, general IQ, experience (the more, the better)
2). Personality and cognitive style
3). Idiosyncratic variations in the weighting of different considerations, selective attention
4). Scales are subjective ('good job' means different things to different people). We can have up to 7 categories. Beyond that, we struggle to assign the correct value
5). Mood (weather, fatigue) - the source of occasional noise
6). Social influence - if a song has been popular in the early stages of the competition, it will likely remain popular later on. Popularity is self-reinforcing
7). Lack of formal procedure for decision making
8). Focusing too much on the outcome of the decision rather than on the process
9). Biases. According to Kahneman, the three major biases that lead to noise are Substitution biases, Conclusion biases, and Excessive coherence.
Substitution biases lead to a misweighting of the evidence: we answer a different question - the one which is easier: For example, if you meet an introverted bald person in glasses, the chances that he is a scientist are pretty low given that scientists as a group represent less than 1%. But if you were asked to make your best guess whether that person was a scientist, you could have assigned a much higher probability. The reason is that you would subconsciously answer the question.
Conclusion biases lead us either to bypass the evidence or to consider it in a distorted way. This often manifests in jumping to conclusions too early when System I is at work (to borrow the term from Kahneman's other bestseller). Experts also suffer from this bias when they look at every fact as evidence of their conclusions and disregard facts that do not support their opinions as irrelevant, not material or an exception.
Excessive coherence magnifies the effect of initial impressions and reduces the impact of contradictory information. We do not like seeing the world as a chaotic, random place. We want to be in control and see events as part of logical chains with cause and effect.
Error in Judgement
Kahneman also discusses measuring a judgement's overall level of error and the two factors that drive it. He uses Gauss's approach called the Mean Squared Errors (MSE). According to the MSE, the best judgement is the one that minimises the mean of squared errors. An error, in this case, is the difference between an estimate made and the actual value. For example, if you have to decide what the temperature will be like tomorrow and you have six different forecasts (e.g. 18, 19, 20, 21, 24, 28), you need to pick the value that would minimise the MSE. In this case, it is 21.7 (when each forecast value is deducted from 21.7, and the result is squared, the mean value of all six values will be the smallest).
Kahneman then states that the error in judgement can be broken down into bias and noise components and suggests a formula:
Overall Error (MSE) = Bias2 + Noise2
Using the formula, Kahneman also makes a crucial point that you improve judgement by reducing any of the two components!
1). Cognitive abilities, general IQ, experience (the more, the better)
2). Personality and cognitive style
3). Idiosyncratic variations in the weighting of different considerations, selective attention
4). Scales are subjective ('good job' means different things to different people). We can have up to 7 categories. Beyond that, we struggle to assign the correct value
5). Mood (weather, fatigue) - the source of occasional noise
6). Social influence - if a song has been popular in the early stages of the competition, it will likely remain popular later on. Popularity is self-reinforcing
7). Lack of formal procedure for decision making
8). Focusing too much on the outcome of the decision rather than on the process
9). Biases. According to Kahneman, the three major biases that lead to noise are Substitution biases, Conclusion biases, and Excessive coherence.
Substitution biases lead to a misweighting of the evidence: we answer a different question - the one which is easier: For example, if you meet an introverted bald person in glasses, the chances that he is a scientist are pretty low given that scientists as a group represent less than 1%. But if you were asked to make your best guess whether that person was a scientist, you could have assigned a much higher probability. The reason is that you would subconsciously answer the question.
Conclusion biases lead us either to bypass the evidence or to consider it in a distorted way. This often manifests in jumping to conclusions too early when System I is at work (to borrow the term from Kahneman's other bestseller). Experts also suffer from this bias when they look at every fact as evidence of their conclusions and disregard facts that do not support their opinions as irrelevant, not material or an exception.
Excessive coherence magnifies the effect of initial impressions and reduces the impact of contradictory information. We do not like seeing the world as a chaotic, random place. We want to be in control and see events as part of logical chains with cause and effect.
Error in Judgement
Kahneman also discusses measuring a judgement's overall level of error and the two factors that drive it. He uses Gauss's approach called the Mean Squared Errors (MSE). According to the MSE, the best judgement is the one that minimises the mean of squared errors. An error, in this case, is the difference between an estimate made and the actual value. For example, if you have to decide what the temperature will be like tomorrow and you have six different forecasts (e.g. 18, 19, 20, 21, 24, 28), you need to pick the value that would minimise the MSE. In this case, it is 21.7 (when each forecast value is deducted from 21.7, and the result is squared, the mean value of all six values will be the smallest).
Kahneman then states that the error in judgement can be broken down into bias and noise components and suggests a formula:
Overall Error (MSE) = Bias2 + Noise2
Using the formula, Kahneman also makes a crucial point that you improve judgement by reducing any of the two components!
Source: Kahneman, Daniel; Sibony, Olivier; Sunstein, Cass R.. Noise: The Flaw in Human Judgement
Using this equation and the MSE concept, Kahneman makes another critical point on reducing error:
"To minimise MSE, you must concentrate on avoiding large errors. If you measure length, for example, the effect of reducing an error from 11cm to 10cm is 21 times as large as the effect of going from an error of 1cm to a perfect hit."
And then he comments on how most people get it wrong following their intuition:
"Unfortunately, people's intuitions in this regard are almost the mirror image of what they should be: people are very keen to get perfect hits and highly sensitive to small errors, but they hardly care at all about the difference between two large errors."
System Noise can be broken down into Level Noise and Pattern noise
Level noise results from the unique characteristics of individuals making judgements (we are all different, some are more optimistic than others).
Pattern noise is the difference between System Noise and Level Noise. It refers to a specific case when the judge deviated from their own average judgement (was more pessimistic or, on the contrary, more optimistic than usual).
An example of a pattern noise is the normally cautious investor who drops his usual caution when shown the plan of an exciting start-up.
Pattern Noise = Stable Pattern Noise + Occasional Noise
Using this equation and the MSE concept, Kahneman makes another critical point on reducing error:
"To minimise MSE, you must concentrate on avoiding large errors. If you measure length, for example, the effect of reducing an error from 11cm to 10cm is 21 times as large as the effect of going from an error of 1cm to a perfect hit."
And then he comments on how most people get it wrong following their intuition:
"Unfortunately, people's intuitions in this regard are almost the mirror image of what they should be: people are very keen to get perfect hits and highly sensitive to small errors, but they hardly care at all about the difference between two large errors."
System Noise can be broken down into Level Noise and Pattern noise
Level noise results from the unique characteristics of individuals making judgements (we are all different, some are more optimistic than others).
Pattern noise is the difference between System Noise and Level Noise. It refers to a specific case when the judge deviated from their own average judgement (was more pessimistic or, on the contrary, more optimistic than usual).
An example of a pattern noise is the normally cautious investor who drops his usual caution when shown the plan of an exciting start-up.
Pattern Noise = Stable Pattern Noise + Occasional Noise
Source: Kahneman, Daniel; Sibony, Olivier; Sunstein, Cass R.. Noise: The Flaw in Human Judgement
Occasion noise (part of the pattern noise) - variability in judgement by the same person depending on particular circumstances (e.g. mood).
Pattern errors arise from a combination of transient (occasion) and permanent factors. The transient factors include those we have described as sources of occasion noise, such as a judge's good mood at the relevant moment or some unfortunate recent occurrence that is currently on the judge's mind. Other factors are more permanent—for example, an employer's unusual enthusiasm for people who attended certain universities or a doctor's unique propensity to recommend hospitalisation for people with pneumonia.
Recurrent vs Singular decisions
Recurrent - estimating the fair value of a stock, hiring personnel, estimating insurance premium
Singular - marriage, buying a house, choosing a job.
The issue with singular decisions is that we cannot measure the degree of noise, but it can easily be present in such situations, just as in recurrent ones (when we can measure the noise).
Hence, the goal of reducing noise is as relevant for singular decisions as it is for recurrent ones.
The issue with noise in a singular decision is that it can violate expectations of fairness and consistency (e.g. the same essay was graded widely differently). It damages the credibility of the system or organisation.
Predictive vs Evaluative decisions
Predictive - decision about the future outcome (e.g. sales growth next year).
Evaluative - promotion, hiring, strategic options for the firm (acquire or develop organically)?
Noise in predictive decisions leads to sub-optimal results (over-spending on a project, achieving considerably lower returns than the market). Noise in evaluative choices can cause the feeling of unfairness and damage the credibility of the organisation.
Occasion noise (part of the pattern noise) - variability in judgement by the same person depending on particular circumstances (e.g. mood).
Pattern errors arise from a combination of transient (occasion) and permanent factors. The transient factors include those we have described as sources of occasion noise, such as a judge's good mood at the relevant moment or some unfortunate recent occurrence that is currently on the judge's mind. Other factors are more permanent—for example, an employer's unusual enthusiasm for people who attended certain universities or a doctor's unique propensity to recommend hospitalisation for people with pneumonia.
Recurrent vs Singular decisions
Recurrent - estimating the fair value of a stock, hiring personnel, estimating insurance premium
Singular - marriage, buying a house, choosing a job.
The issue with singular decisions is that we cannot measure the degree of noise, but it can easily be present in such situations, just as in recurrent ones (when we can measure the noise).
Hence, the goal of reducing noise is as relevant for singular decisions as it is for recurrent ones.
The issue with noise in a singular decision is that it can violate expectations of fairness and consistency (e.g. the same essay was graded widely differently). It damages the credibility of the system or organisation.
Predictive vs Evaluative decisions
Predictive - decision about the future outcome (e.g. sales growth next year).
Evaluative - promotion, hiring, strategic options for the firm (acquire or develop organically)?
Noise in predictive decisions leads to sub-optimal results (over-spending on a project, achieving considerably lower returns than the market). Noise in evaluative choices can cause the feeling of unfairness and damage the credibility of the organisation.
Part III - Actionable points on how to improve your decision making
1. Check your mental state
Ask yourself if there is anything that makes you overly excited or too pessimistic, an event that you keep regretting or other factors making you feel frustrated or nervous, or experience any other feelings keeping you out of emotional balance. Lack of quality sleep also impacts your decision-making skills.
2. Defer the decision, do not make an immediate decision
Returning to the same question in a few days may not lead you to find a better answer, but a combination of your first thought and the second one would lead to a better final decision. Extending your decision-making process reduces the impact of random factors (including your emotions). When you slow down before making a decision, you switch on your System 2, which is more analytical and critical thinking. You will increase the chances of considering more relevant points and getting a more rounded view before deciding.
3. Work out guidelines/checklists/rules
In investing, for example, this can be the minimum number of profitable years a company should have for you to invest. Or as simple as never investing in an IPO, or a minimum level of historical growth rate, specific leverage criteria etc. Rules also help distinguish between objective facts and personal preferences and values. We should try to see the world as it is, not as we want it to be (e.g. a typical mistake is a reaction like - "It doesn't make any sense, this cannot happen."). Rules, formulas and algorithms are better than human judgement not because of some superior insight they bring but because they are noiseless.
The primary reason model does better than people is that they are less noisy. They may not reflect reality perfectly, and their predictions will not be 100%, but they will still deliver better results than humans whose decisions are too noisy.
Guidelines are better than rules as they do not entirely eliminate the need for judgements. Yet guidelines reduce noise because they decompose a complex decision into several easier sub-questions.
A formula does not have to be complex to be helpful; a simple one is also good. Complex rules will often give you only the illusion of validity and, in fact, harm the quality of your judgments.
I think using more equal weights in investment portfolios will reduce the number of decisions you make and potentially improve your returns. It will force you to think about each investment and its characteristics and will not give you an escape route when you discover increased risk by saying, 'This is just a 2% position for me. I cannot lose too much'. As a compromise, you may have only two or three types of weights (e.g. 20% for core stocks where the risks are extremely low, like Berkshire Hathaway; 10% for all others and 5% for new positions to have the opportunity to increase their weight to 10% gradually and, thus, reducing timing risk).
The primary reason model does better than people is that they are less noisy. They may not reflect reality perfectly, and their predictions will not be 100%, but they will still deliver better results than humans whose decisions are too noisy.
Guidelines are better than rules as they do not entirely eliminate the need for judgements. Yet guidelines reduce noise because they decompose a complex decision into several easier sub-questions.
A formula does not have to be complex to be helpful; a simple one is also good. Complex rules will often give you only the illusion of validity and, in fact, harm the quality of your judgments.
I think using more equal weights in investment portfolios will reduce the number of decisions you make and potentially improve your returns. It will force you to think about each investment and its characteristics and will not give you an escape route when you discover increased risk by saying, 'This is just a 2% position for me. I cannot lose too much'. As a compromise, you may have only two or three types of weights (e.g. 20% for core stocks where the risks are extremely low, like Berkshire Hathaway; 10% for all others and 5% for new positions to have the opportunity to increase their weight to 10% gradually and, thus, reducing timing risk).
4. Wisdom of the crowd
A group of people making independent judgements will produce a better outcome than an individual, even an expert. A group of independent experts will generate even better results. There are also decisions made at group meetings. It is then preferable to have private votes first and then the discussion. If you start the conversation first, the order in which each group member speaks can impact the thinking of other members who are yet to speak. The order in which experts talk at a meeting can influence the final decision made by the group.
5. Try to find your own 'Charlie Munger'
Warren Buffett praised his long-life business partner and friend, Charlie Munger, for helping him make better decisions. Kahneman refers to studies suggesting that a second expert's view will significantly improve your decisions. If you ask the same question yourself at a later time (Point 2), you will only receive a third of the benefit compared to asking your expert friend.
6. Watch out for great stories
It makes a case look more coherent, you become overconfident and may pay attention to irrelevant facts (but fit well the overall story). Going back to Point 3, if you have a checklist, you know what you are looking for and do not let random facts impact your views and start building an exciting story.
In investing, rather than reading an in-depth research note on a stock or checking a company's PE multiple, it is better to first look at the key fundamentals like sales growth, profit margins, FCF and dividend payments, and ROIC. Afterwards, think about the fair price for such a business. Then check the actual price and multiple. Think about the reasons why the market may have a different view than you. Ted Weschler also described this method in one of his rare interviews in 2022.
In investing, rather than reading an in-depth research note on a stock or checking a company's PE multiple, it is better to first look at the key fundamentals like sales growth, profit margins, FCF and dividend payments, and ROIC. Afterwards, think about the fair price for such a business. Then check the actual price and multiple. Think about the reasons why the market may have a different view than you. Ted Weschler also described this method in one of his rare interviews in 2022.
7. Keep a diary to avoid a hindsight bias
Write down the rationale for each decision, the factors behind it, and your concerns at that time.
8. Try to take more outside view, less inside view
Statistical thinking [outside view] considers individual cases as instances of broader categories [base rate]. A specific case is not seen as resulting from a chain of particular events but is viewed as a statistically likely (or unlikely) outcome, given prior observations of cases that share predictive characteristics with a case in point. The inside view is looking at all the specific details of a particular case and trying to work out the decision from the inside. The risk with the inside view is that you overweight specific facts and draw wrong conclusions ignoring the bigger picture and other points you may not be aware of.
9. Think about the opposite decision (instead of buying a stock, you short it), then write down why this could be a good decision
This forces you to consider new information and factors for the first time. If the opposite decision is correct, what reasons could be behind you making the wrong decision (the original one)? What assumptions could have been incorrect? What does it imply? Was your first decision too optimistic or too conservative? After going through the exercise, make a new estimate (judgement). This technique is partially an alternative to asking a second opinion as you effectively become that other person when you think on an opposite view.
It is also helpful to seek counterarguments and speak to people who may disagree with you or have different perspectives.
It is also helpful to seek counterarguments and speak to people who may disagree with you or have different perspectives.
10. Decompose a complex question into sub-components
In investing, it means having a clear set of criteria for an ideal investment and then analysing each component of a company in question. It is a better way than trying to decide straight away if this is a great company to invest in.
11. To improve decisions at large firms, it is vital to have the right culture where disagreement is accepted and different views are encouraged
According to Kahneman, the most significant factor for noise at large firms is "simply the discomfort of disagreement. Most organisations prefer consensus and harmony over dissent and conflict. The procedures in place often seem expressly designed to minimise the frequency of exposure to actual disagreements and, when such disagreements happen, to explain them away."
This point is similar to Dalio's principle of 'radical transparency'.
This point is similar to Dalio's principle of 'radical transparency'.
12. Ranking is better than ratings. In investing, it means that rather than estimating a company's fair value precisely, it is better to pick the best investment out of several options
The more alternatives you have, the better. This also explains why generalists are generally better than narrow-sector specialists.
Explicit comparisons between objects of judgment support much finer discriminations than ratings of objects evaluated one at a time. When analysing a company, it helps if you can relate it to other companies rather than try to come up with the most precise valuation. Use comparative judgements (ranking) to reduce noise.
Studying many companies helps to evaluate new investment opportunities better. Not surprisingly, one of the best investors in the world, Warren Buffett, developed a habit of going through Value Lines stock snapshots in the early years of his career.
Generalists also have an advantage as they have more sectors and a wider range of companies to put a particular investment opportunity in the right context. Narrow sector-focused experts will miss potentially more lucrative opportunities or bigger trends (e.g. buying an oil stock that trades on a single PE at a time of record oil prices and emerging signs of recession).
Explicit comparisons between objects of judgment support much finer discriminations than ratings of objects evaluated one at a time. When analysing a company, it helps if you can relate it to other companies rather than try to come up with the most precise valuation. Use comparative judgements (ranking) to reduce noise.
Studying many companies helps to evaluate new investment opportunities better. Not surprisingly, one of the best investors in the world, Warren Buffett, developed a habit of going through Value Lines stock snapshots in the early years of his career.
Generalists also have an advantage as they have more sectors and a wider range of companies to put a particular investment opportunity in the right context. Narrow sector-focused experts will miss potentially more lucrative opportunities or bigger trends (e.g. buying an oil stock that trades on a single PE at a time of record oil prices and emerging signs of recession).
13. Avoid big mistakes (e.g. buying an overleveraged company). They reduce noise much more than small mistakes
"To minimise MSE, you must concentrate on avoiding large errors. If you measure length, for example, the effect of reducing an error from 11cm to 10cm is 21 times as large as the effect of going from an error of 1cm to a perfect hit." "Unfortunately, people's intuitions in this regard are almost the mirror image of what they should be: people are very keen to get perfect hits and highly sensitive to small errors, but they hardly care at all about the difference between two large errors."
14. Correcting predictions
I. Make your intuitive guess. (A)
II. Look for the average (mean) for such cases. Forget about individual characteristics. (B)
III. Estimate the value of the information you have (quite challenging). It should be expressed as a correlation between the evidence and the outcome you try to predict. In the social sciences, correlations of more than .50 are very rare. Many correlations that we recognise as meaningful are in the 0.20 range. (C)
IV. The final step is a simple arithmetic combination of the three numbers you have now produced. Expressed as a formula, the Final outcome = Intuitive guess X Correlation + Mean (Base Rate) x (1 - Correlation). Or, Final outcome = A x C + B x (1 - C). If rearranged, the Final outcome = B + (A - B) x C.
Suppose you are trying to estimate the growth rate of a company ten years from now. So far, it has averaged 25% CAGR over the past 15 years. Given your knowledge of the product, management and general understanding of the competition, you think the company can maintain the same rate in Year 10. Then you should look for the base rate. Let us assume that a typical company can grow its sales at about a 5% rate, which roughly matches normal GDP growth of 2% and some inflation (let's say 3%).
Then you have to estimate the quality of the information you have for your specific company. What is the correlation between your data and the future outcome (sales growth of 25% in ten years)? Given that 0.5 is considered very high and unless you have unique insights and skills, a 0.25 correlation would be viewed as good.
The final step is to work out the corrected prediction using the formula above. The adjusted growth rate is only 10%, much closer to an average company than to the historical rate of the company you are analysing. This is, of course, a very theoretical exercise. Your experience, skills, and knowledge of the company, sector and other circumstances can dramatically impact the forecast.
One final and pretty sad thought about this formula is that you will most often reject investing in startups and the next Amazon or Google, given that statistics work against such companies (the base rate of success is extremely low). And identifying such companies using financial statements or general knowledge about the product cannot materially boost your predictive power (correlation).
II. Look for the average (mean) for such cases. Forget about individual characteristics. (B)
III. Estimate the value of the information you have (quite challenging). It should be expressed as a correlation between the evidence and the outcome you try to predict. In the social sciences, correlations of more than .50 are very rare. Many correlations that we recognise as meaningful are in the 0.20 range. (C)
IV. The final step is a simple arithmetic combination of the three numbers you have now produced. Expressed as a formula, the Final outcome = Intuitive guess X Correlation + Mean (Base Rate) x (1 - Correlation). Or, Final outcome = A x C + B x (1 - C). If rearranged, the Final outcome = B + (A - B) x C.
Suppose you are trying to estimate the growth rate of a company ten years from now. So far, it has averaged 25% CAGR over the past 15 years. Given your knowledge of the product, management and general understanding of the competition, you think the company can maintain the same rate in Year 10. Then you should look for the base rate. Let us assume that a typical company can grow its sales at about a 5% rate, which roughly matches normal GDP growth of 2% and some inflation (let's say 3%).
Then you have to estimate the quality of the information you have for your specific company. What is the correlation between your data and the future outcome (sales growth of 25% in ten years)? Given that 0.5 is considered very high and unless you have unique insights and skills, a 0.25 correlation would be viewed as good.
The final step is to work out the corrected prediction using the formula above. The adjusted growth rate is only 10%, much closer to an average company than to the historical rate of the company you are analysing. This is, of course, a very theoretical exercise. Your experience, skills, and knowledge of the company, sector and other circumstances can dramatically impact the forecast.
One final and pretty sad thought about this formula is that you will most often reject investing in startups and the next Amazon or Google, given that statistics work against such companies (the base rate of success is extremely low). And identifying such companies using financial statements or general knowledge about the product cannot materially boost your predictive power (correlation).