We have come a long way in terms of analyzing the financial markets. The evolution of theories pertaining to the markets dates back to the 17th century where the famous tulip mania term was coined when a bubble in an economy was first identified. And now in the 21st century, we analyze financial markets through the lens of advanced behavioral finance theories.
Going back to the 1970s, the efficient markets hypothesis (EMH) was at the height of its dominance and was assumed to be proven beyond doubt. The idea emerging was that speculative asset prices such as stock prices always incorporate the best information about fundamental values and that prices change only because of good, sensible information meshed very well with theoretical trends of the time. Shiller (2003).
However, Eugeme Fama (1970) mentioned in his report that market efficiency theory has some anomalies like serial dependencies in stock market returns. In the 1980s, the consistency of the efficient market hypothesis was first tested against the light of excess volatility. But the stock prices had more volatility than an efficient markets hypothesis could explain, which questions the basic underpinnings of the entire hypothesis as most of the volatility was unexplained.
In the 1990s, the spotlight moved away from econometric analysis toward the human psychology models related to financial markets or so-called behavioral finance. According to Shiller (2003), Feedback models, one of the oldest theories about financial markets translated as; When speculative prices go up, creating successes for some investors, this may attract public attention, promote word-of-mouth enthusiasm, and heighten expectations for further price increases. The feedback may be an essential source of much of the apparently inexplicable randomness that we see in financial market prices.
The efficient markets theory, as it is commonly expressed, asserts that when irrational optimists buy a stock, smart money sells, and when irrational pessimists sell a stock, smart money buys, thereby eliminating the effect of the irrational traders on market price. But finance theory does not necessarily imply that smart money succeeds in fully offsetting the impact of ordinary investors. In recent years, research in behavioral finance has shed some important light on the implications of the presence of these two classes of investors for theory and also on some characteristics of the people in the two classes.
Of the various biases studied under behavioral finance, the three biases that we have focused on are anchoring bias, confirmation bias, and overconfidence bias. We have analyzed 4 papers, each covering different biases in-depth, with the last one examining two of the biases. After examining the findings of the research papers, we conclude by commenting on the intertwined relationship of the biases and how we can resolve them.
Of the several systematic biases noted by Tversky and Kahneman (1974) causing large and predictable forecast errors, anchoring bias is one of them, wherein there is a tendency to attach our thoughts to a reference point – even if there is no logical relevance to the same. They define anchoring to occur when “people make estimates by starting from an initial value that is adjusted to yield the final answer… adjustments are typically insufficient… different starting points yield different estimates, which are biased towards the initial values”
One of the experiments Tversky and Kahneman employed involved half of the subjects estimating the value of 1x2x3x4x5x6x7x8 and the other half estimating 8x7x6x5x4x3x2x1 within 5 seconds. The average answers from the 2 groups were 512 and 2250 respectively. In this case, it can be inferred that the starting number of the series was a reference point and had a significant impact on the estimate even though the answer is the same in both cases. While making economic forecasts, the tendency to underweight recent information can cause forecast errors.
In their paper, ‘Anchoring bias in consensus forecasts and its effect on market prices, Sean D. Campbell and Steven A. Sharpe investigate surveys done by Money Market Services (MMS), which collect expert consensus forecasts between 1991 and 2006 and are widely used. Then, the focus turns toward monthly macroeconomic data releases, which are previously shown to have a significant impact on market interest rates.
Keeping in line with early studies on anchoring bias, macroeconomic forecasts have been tested to show if they have the properties of rational expectations. This has been done, by running a regression of actual values of data releases (dependent variable) on the recent forecast (independent variable). The researchers run an alternative of this basic rationality test, using “surprise” (difference of actual and forecast) values regressed against the previous month’s forecasts.
The equation is revised further and finally, we check if surprise can be regressed meaningfully upon the lagged values of the difference of forecast and the average values of past “h’’ months. If the slope coefficient of the independent variable is > 1, it would imply that the forecasts are systematically biased towards the lagged values.
Earlier studies have shown the substantial impact of news data releases on financial market prices. The additional study being done here is to check whether market prices react more to the predictable component or the residual component of the surprise. The predictable component has been calculated as the forecasted component of surprise as given by the OLS technique. If the predictable and residual components of surprise are equal, this means the market can see through the anchoring bias and the impact should be lower to that extent.
The 2-year and 10-year Treasury yields have been taken as a proxy for prices. The data releases which have previously shown statistically significant impact on prices (Consumer Confidence, Consumer Price Index (CPI), Core CPI, Durable Goods Orders, Industrial Production, ISM Manufacturing Index, New Home Sales, Retail Sales, Retail Sales (ex-auto)) have been included as independent variables.
The findings in the MMS survey include mean surprise being unconditionally unbiased, the standard deviation of forecasts being lower than that of actuals, and a statistically significant degree of negative serial correlation for surprise values. This suggests the anchoring of forecasts on the most recent value of the lagged variable.
Regarding the data releases, the interest rate reaction has been taken as the difference between the quote 5 minutes before the release and 10 minutes after the release. The important finding here is that the degree of serial correlation is statistically significant for 3 of the 10 releases, indicating that the interest rate responses might be partly predictable. Also, the one-month anchoring along with the three-month anchoring model shows a significantly positive coefficient for 8 out of 10 releases. This is true for time-series data as well, with the degree of persistence high.
Not only is anchoring widely prevalent, the size is also substantial. For instance, forecasters put 40% weight on one month-lagged value and 60% on expected value for Consumer Confidence data release.
Other major finding is that the unexpected component of surprise on yields is significant for 9 out of 10 releases but the market participants do not react significantly to predictable components. Thus, some of them do not take such forecasts at face value but discard the anchoring bias estimate from the forecast.
Still, it is inappropriate to judge forecasters as irrational as there are other considerations such as faulty incentive structure which prefers common view to own private signal or issuing more conservative forecasts than warranted to minimize chances of being wrong. It is also possible that the forecaster may be viewing a forecast as a part of the underlying trend which is reflected in a lower standard deviation (smoothening).
Confirmation bias, a term coined by Peter Wason in 1960 is the tendency of seeking or interpreting facts partially to existing beliefs and one’s own expectations while testing a hypothesis.
In the paper, ‘Confirmation Bias: A Ubiquitous Phenomenon in Many Guises’, Raymond S. Nickerson has firstly reviewed experimental evidence of confirmation bias. Secondly, real-life practical examples of confirmation biases have been laid out. The third section notes different possible explanations of the bias proposed by researchers over the years. And in the fourth section, he has addressed the question of the effects of confirmation bias and its underlying utility.
1. Experimental Evidence:
Empirical evidence suggests that Confirmation bias is strong and extensive and is present in various forms. The results of the experiments also suggest that once a stand has been taken by a person, defending that stand becomes of primary importance to the person who made the decision. Despite one having weighed both sides evenly in the first place, this phenomenon is still present when one is provided with more facts to test the hypothesis
Hypothesis-Determined Information Seeking and Interpretation:
One fails to seek information that would tend to counter the initial stand and thereby fails to consider the alternate hypothesis in the Bayesian framework leading to inaccurate calculation of likelihood ratios which represents the ratio of two conditional probabilities testing the truthfulness of each hypothesis in a given scenario. (Doherty, Mynatt, Tweney, & Schiavo, 1979; Griffin & Tversky, 1992)
Looking only or primarily for positive cases despite not having vested interests:
One would often look for examples that would be classified as illustrations of the sought-for concept if the hypothesis to be tested were correct. Studies demonstrating selective testing done by Wason (1960) based on the selection of triplets wherein People typically tested hypothesized rules by producing only triplets that were consistent with them and thereby precluded themselves from discovering that they were inaccurate in choosing the test cases.
Overweighting positive confirmatory instances:
People tend to overweight positive confirmatory response and underweight negative confirmatory response. This asymmetry is because of the confidence one has in its initial preference. The need for accuracy as one important determinant of hypothesis-evaluating behavior along with several motivational factors such as self-esteem, control, and cognitive consistency.
One may associate confirmation bias with the perseverance of false beliefs. However, it is independent of the truthfulness of the falsity of the underlying belief. People also tend to express a higher degree of confidence than is rationally acceptable on the views initially formed. Being forced to evaluate the contradictory viewpoint wherein reasons for them are asked to think of, has reduced overconfidence in some instances. Another reason why people tend to be overconfident of their knowledge is that once a person fixes his mind on one alternative, he is more inclined to think for the reasons behind it and fail to think about possible alternatives.
2. The Confirmation Bias in Real-World Context
Once a policy has been adopted and implemented in any country, all efforts are made to defend them rather than analyzing any other possible alternatives. This same reason kept the US engaged in war with Vietnam for more than 17 years.
Knowledge was a barrier to knowledge. For many years commonly accepted principles in medicines that were formed due to mere observations were accepted without fully analyzing and testing the. Modern science has brought changes in a lot of practices based on hypothesis testing in absence of any confirmatory bias.
Jurors are expected to restrain from forming any sorts of judgments and maintain open minds during the deliberation phase which otherwise could lead to them forming an early opinion and later leading to focusing only on the facts that support their earlier claims leading to confirmatory bias. This ensures that they evaluate each fact and evidence carefully without any preconceived notions.
Both the common man and the scientists have their own theories based on observations and they are both tending to fall prey to confirmation bias by paying attention to facts that could agree or not with their own conjecture. For years scientists have not accepted each other’s theories and have tried their best to come up with an alternative explanations to various things in nature. Galileo did not accept Kepler’s hypothesis of the moon being responsible for tidal currents. Newton rejected the theory that could be over 6,000 years old. Huygens and Leibniz rejected Newton’s concept of universal gravity.
3. Explanations of the Confirmation Bias
Several reasons can be attributed to explain the existence of confirmation bias such as ego, Cognitive limitations, lack of understanding of logic, or some fundamental value. However, they can be broadly explained using the following theories as described in the paper
Desire to believe:
Researchers have observed that people find it easier to believe propositions they would like to be positive/truth than propositions they would like to be negative/lie. There is a positive correlation between what one would consider being true and what is desired. And these beliefs are created basis the preferences people have developed over a period. The fact that we normally discount facts that count our belief is witness to the importance we put on them. Motivation (a strong wish to confirm) will lead to confirmation bias, but cognitive determinants such as our beliefs determine its magnitude.
Information-Processing Bases for Confirmation Bias:
People are fundamentally constrained by their minds in a way that only one hypothesis or only one side occupies the mind and the mind is incapable of processing its alternatives leading to an information processing base for confirmation bias.
Positive-Test Strategy or Positivity Bias or Congruence heuristic:
In absence of any other evidence, one is more inclined toward finding evidence that may prove the hypothesis to be true rather than false. One is less likely to attempt in rejecting the hypothesis and considering them as false than attempting to prove the hypothesis as truth.
And generally, from the hypothesis statement, it is easy to make out which statement is positive, and which one is negative, and one is normally more inclined toward proving the positive statement true.
Conditional Reference Frames:
When one is asked to reason as to why a hypothesis may be true, they are already somewhat convinced that the hypothesis is true Koehler found out that finding a reason is not of importance in this case and simply by arriving at a focal hypothesis one can be convinced that the hypothesis is in fact true. Calling attention to a desired hypothesis leads to one believing it to the focal hypothesis which in turn leads to the acknowledgment of the conditional reference frame wherein the focal hypothesis is accepted to be true.
Pragmatism and Error Avoidance:
When the consequences of treating a true hypothesis to be false are far more than that of treating a false hypothesis to be true it may lead to confirmation bias dictated by some normative models of reasoning and common sense. In this case, people may be more inclined towards seeking small rewards rather than having to go through a costly consequence but not on the pure objective of analyzing and testing the pure hypothesis.
At every level of education, importance is placed on having reasons for what we believe in. If one is compulsorily forced to practice finding reasons for one’s own belief rather than finding alternatives to our beliefs, then one is ought to hard wire confirmatory bias in his or her life. A standard method for teaching places emphasis on giving supporting evidence to strengthen our beliefs rather then countering it or presenting alternating beliefs that may be possible.
4. Utility of Confirmation Bias
Utility in Science-principle of fallibility:
Hypothesis are made stronger when highly competent scientists attempt to prove them wrong and fail at it rather than some moderate scientists trying to prove the hypothesis to be correct. To the extent that such a belief is accurate, an attempt should be made to test any new scientific theory by trying to negate it and proving it wrong. Secondly, it may be possible to hold a belief for justified reasons without being able to produce concrete evidence for the same. Also, in certain cases conservatism may lead to confirmation bias and a lot of them have led to scientific discoveries.
Focused and single-mindedness to overcome conflicts-
Although vague this theory aims to suggest that by not overanalyzing and thinking about infinite alternatives to a hypothesis it is more beneficial in certain cases to have a confirmatory bias and satisfy one’s own ego. Precisely these qualities permitted the 17th-century New England Puritans to establish a society with the ingredients necessary for survival and prosperity which might have otherwise they might have fallen to the unpredictable perils of the wilderness.
Overconfidence typically refers to an irrational and exaggerated belief in one’s abilities with respect to successfully solving any problem/task. In a finance context, it is used for forecasting events.
Overconfidence bias leads to the false assumption that someone is better than others, due to their own false sense of skill, talent, or self-belief.
Understanding where the direction of market fluctuation is one of the most important skills in finance and investing. In this industry, most professionals think they are above average when logically it is impossible.
In the publication ‘Self-Serving Attribution Bias, Overconfidence, and the Issuance of Management Forecasts’ by Robert Libby and Kristina Rennekamp, they conducted an abstract experiment and a survey of experienced financial managers in terms of issuing earnings forecasts.
The researchers primarily wanted to examine whether the managers engage in self-serving attribution which in turn increases overconfidence. Self-serving attribution is defined as a tendency to attribute positive outcomes to their own internal skill-set and negative outcomes to external factors.
The research paper focused on the hypothesized causal relationships to arrive at the conclusion. Following were the hypotheses:
- Higher ratings of first-round performance will be associated with self-serving attribution
- Self-serving attribution resulting from favorable perceptions of first-round performance increases confidence that second-round performance will exceed first-round performance
- Participants who are higher in stable individual measures of overconfidence will also be more confident that their second-round performance will exceed first-round performance
- Participants that are more confident in their second-round performance will be more likely to commit to doing better in the second round than in the first round
The experiment involved 57 participants who were doing MBA from Cornell University. The experiment was designed as a two-round test consisting of trivia of 25 questions of mixed difficulty levels. There were two conditions in the tests; a low-difficulty scenario consisting of 15 easy questions, 5 moderate, and 5 hard questions. Whereas a high-difficulty scenario consists of 5 easy questions, 5 moderate, and 15 hard questions. The participants were incentivized with $2 if they answered correctly. After 1st round, they were shown how many they answered correctly but they never knew the difficulty level. After this, they were asked to estimate the difficulty levels of the questions. Further, they were asked to gauge their performance in terms of internal and external factors. The internal factors were “skill” and “effort” whereas external factors were “luck” and “difficulty”. The result of the experiment was such that those who performed bad they attributed it to the external factors whereas those who performed well they attributed it to the internal factors.
The experiment was taken into the next stage where a confidence trait was developed among the high performers to perform well in the future round as well. For this round, the participants knew the difficulty mix in advance. Although the difficulty level was the same in both rounds, participants never knew about it since they were unaware of the difficulty level in the 1st round. They were told that will be incentivized with $2 per correct question before committing for the test. Then their confidence level was tested in terms of appearing for the 2nd round. After that were told that if they commit to improving their performance in the 2nd round with respect to the 1st round then they will gain $2.5 for the correct answer or lose $1.5 for the wrong answer. The last part of the experiment involved a true test and without any manipulation to gauge the overconfidence trait in the managers when they were asked to collect their payments a week later. The whole experiment was designed in such a way that it captures the variations in the real world that managers always face. In a nutshell, the experiment gauges the self-attribution, and overconfidence levels throughout the test which involved forecasting their performance.
The results of the experiments proved the hypotheses that:
- Those who got a low-difficulty set scored higher than the higher-difficulty guys in both rounds.
- Those who scored well attributed their performance to internal factors and those who performed poorly attributed to external factors.
- After 1st round, those who did well believed that it was because of internal factors and they are less likely to improve their 2nd round performance
- It was also observed that those who had stable confidence levels, they also expected that they will do better in the 2nd round.
- Those who were overconfident and committed that they will do better as compared to those who didn’t commit, actually didn’t have much difference in terms of their payouts and in fact made optimistic forecasting errors.
The primary evidence was that managerial overconfidence has contributed in making that optimistic forecasting decision. The experiment also shows that participants engage in self-serving attribution, giving greater weight to internal than external factors as explanations for good performance. This in turn increased confidence and expectations of improved future performance in the managers but which resulted in more forecasting errors because of overconfidence and self-serving attribution bias.
Confirmation bias and Overconfidence bias
Let us look into the above-combined effect with the help the research paper on ‘Information Valuation and Confirmation Bias in Virtual Communities: Evidence from Stock Message Boards’ by JaeHong Park.
As part of this research paper, a study was carried out on a message board website to test the visitors for confirmation biases and their inability to rationally evaluate the new information they come across.
As part of this study, they had evaluated five hypotheses namely
- a) Subjects with stronger initial beliefs about future performances of stock were more likely to display bias
- b) Subjects with higher perceived knowledge were more likely to show confirmation bias c) Bias will be low when both initial sentiment and perceived knowledge is low
- d) Gap between forecasted return and actual return will be wider for those who display biases
- e) the Higher extent of bias will lead to greater trading frequency.
As a result of their testing it was revealed that all of their initial hypothesis held true at 1% significance level. Strength of belief had a positive co-efficient of 0.214, Perceived knowledge had a positive co-efficient of 0.268, Combination of both had a negative co-efficient of 0.178, Return gap had a negative co-efficient of 0.147, and trading frequency had a positive coefficient of 0.159.
As a result of this study, it was revealed that the website visitors did display confirmation bias which enhanced their overconfidence and boosted their optimism levels where they relied heavily on their expected returns. Thus, the overall perceived advantage of such message boards which was initially deemed to be a source of information for retail investors was put to question.
The confirmation bias in information valuation was an important determinant for investor overconfidence. The stronger the belief the more was the tendency to seek information confirming their beliefs and more was the degree of overconfidence.
This overconfidence underestimated the volatility of random events in financial markets which led to an increasing difference of opinions among investors and led to higher trading frequencies.
Overconfidence is reflected in the illusion of control over random events, Illusion of knowledge, and self-attribution bias. These in turn encourage investors to trading more even though this may lead to worse performances.
On account of these biases, individual investors are motivated to trade more frequently and experience anchoring to their expected returns but end up with poor performance as higher trade frequency leads to greater transaction costs.
The fundamental reason for investors to experience confirmation bias leading to all other biases is the discomfort they experience when presented with contrary information. The stronger the initial belief of future stock performance the most discomfort that an individual would experience to process contrary information.
According to self-enhancement account, people are motivated to hold positive views of themselves and their future and thus value confirmatory information more because it enhances their perceived sense of knowledge about relevant information.
Investors who perceived themselves to be knowledgeable also back their initial decision and experience discomfort to process contrary information. On a contrary a person who starts evaluation without any sentiments and lower perceived knowledge would be more interested in gathering information will strive to evaluate each piece of information rationally would fit into the Bayesian model thereby avoiding the biases trap.
Once inside the trap of biases an investor will experience overconfidence and anchor himself to his expected return as a result of which he will have increased trading frequency and realize greater deviation of returns from his initial expected returns.
It is important to note that the individual biases observed above do not work in insolation but in tandem, it is the magnified effect of the above biases working together which moves an Individual away from exhibiting rational decision making by using the Bayesian framework for new information which they come across.
As an application of this in everyday life, we find repeated references by market experts saying Nifty at 10,000 would be a strong support level…If Nifty breaks, 9750 would be the next strong support and so on.
People working to identify such support and resistance levels, are likely to make them as a reference point for buying and selling, in the hope that they can leverage what others think in the market.
They may disregard contrarian evidence (confirmation bias) and keep the trade open, overconfident that they would be right. This remains until there is massive volume on the other side of the trade and hence, he is left with no choice but to book a heavy loss.
The cycle continues as he searches for the next stock that recoup his losses in the earlier trade.
Thus, we see that the 3 biases we analyzed may be intertwined in many cases and one may lead to another.
To resolve the biases, having an objective approach (in the sense that opinions have to be backed by numbers) helps reducing the importance of reference points. Also, agility in reacting to new information pertaining to one’s own investments can prevent formation of reference points.
Another must-have is dissenting opinions from credible sources which helps remove any emotions associated with the investment.
Humility in understanding that even the best investors make mistakes and accepting one’s mistakes is key. Learning from our own as well as others’ mistakes can reduce overconfidence.
- Robert J. Shiller (Winter, 2003). From Efficient Markets Theory to Behavioral Finance. The Journal of Economic Perspectives
- Sean D. Campbell and Steven A. Sharpe (April 2009). Anchoring Bias in Consensus Forecasts and Its Effect on Market Prices. The Journal of Financial and Quantitative Analysis
- Raymond S. Nickerson (1998). Confirmation Bias: A Ubiquitous Phenomenon in Many Guises. Review of General Psychology
- Robert Libby And Kristina Rennekamp (2011). Self-Serving Attribution Bias, Overconfidence, and the Issuance of Management Forecasts. Journal of Accounting Research
- JaeHong Park, P. K. (December 2013). Information Valuation and Confirmation Bias in Virtual Communities: Evidence from Stock Message Boards. Information Systems Research, 1050 – 1067.