experimental and neuro finance
play

Experimental and Neuro Finance Elena Asparouhova (U Utah) and - PDF document

7/26/12& Experimental and Neuro Finance Elena Asparouhova (U Utah) and Peter Bossaerts (Caltech) Melbourne, July 25, 2012 what we do We study financial decision making, all the way from the level of markets (asset pricing) down


  1. 7/26/12& Experimental and Neuro Finance Elena Asparouhova (U Utah) and Peter Bossaerts (Caltech) Melbourne, July 25, 2012 what we do • We study financial decision making, all the way from the level of markets (“asset pricing”) down to the individual (“behavioral finance”). • We don’t just want to describe (GARCH, Prospect Theory,…); we want to understand! • Why do prices move to levels where only systematic risk is priced? • Why are humans able to track risk – but confused about outliers (black swans)? • Our methodology: Experiments • Observe humans make (financial) decisions and interact in a controlled setting 2 1&

  2. 7/26/12& Experiments!? • We eventually want to understand real-world financial markets and the creatures that inhabit them (humans). • Are these markets not “too big” and “too complex” to be studied in the laboratory? • Too complex? That is precisely their problem… They cannot be described in terms of simple equations; so you need control, i.e., laboratory study • Too big! Sure, but we have to start somewhere. Without experimentation, we are "likely to go completely astray into imaginary conjecture” [Hannes Alfven, Nobel (Astro)Physics] 3 Analogy with astrophysics The TORPEX at EPFL, a tiny vacuum- like space where physicists generate “plasma” using a microwave “oven” in order to study physical processes inside stars (that float in a real vacuum). It’s small relative to the real stars, but the only way to be sure that we correctly interpret the “signals” that the stars send to us! 4 2&

  3. 7/26/12& More about experiments • Asset pricing theory in particular is not ready for real markets; it is too stylized, it leaves out too many details (taxes, intermediaries, ambiguity about returns, demographics, politics…) • Yet it is ready for laboratory experiments (we hope to convince you of this today): it makes precise predictions about prices and allocations, but still leaves out the details of how to get there • Experiments lead to a deeper understanding of the theory ( Richard Feynman , Nobel Prize Winner) • … and to new theory (in our case: about HOW markets equilibrate) 5 Two examples • The CAPM • Risk prediction in the human brain 6 3&

  4. 7/26/12& CAPM • Theory: In equilibrium, the expected return on risky securities is solely determined by their covariance with aggregate risk (“beta”) • Equivalent: the “market portfolio” will give you maximum expected reward for its risk (risk=return variance), or the Sharpe ratio of the market portfolio is maximal • Sharpe ratio = Expected return on portfolio minus riskfree rate / return standard deviation 7 Lab • Anywhere from 20 to 30 subjects in the lab (up to 70 if participating from home). • The participants are given initial allocations of securities (called, say, A, B and Notes) and cash. • Initial allocations are risky and different for different subjects, and they can be improved upon through trading (with other subjects). 8 4&

  5. 7/26/12& LAB • Each laboratory session is a repetition of the same situation, and each repetition is called a period. 6-10 periods in a session. • Period=Initial endowments, trading, final portfolios, payoff. • Example: • Three markets (one riskfree, shortsales allowed), several pe- riods. • Three states; determine liquidation value at end of each pe- riod; known probabilities. Security State X Y Z A 170 370 150 B 160 190 250 Notes 100 100 100 9 LAB • Endowment of risky securities and cash, refreshed every pe- riod. E.g., 5 of A, 4 of B, and 400 cash (may vary across subjects). • Loan repayment of, say, 1900 at end of period (leverage!). • Trade through a web-based open book system, Marketscape , developed at Caltech. Example Of This Experiment: 011126 Draw Subject Signup Endowments Cash Loan Exchange Type Type Reward A B Notes Rate (#) (franc) (franc) (franc) $/franc D 18 125 5 4 0 400 2200 0.04 18 125 2 8 0 400 2310 0.04 5&

  6. 7/26/12& Logout Exit Market Marketplace closes in 01:56:25 Navigation Markets -- Actual Cash on hand: $500.0 Marketplace A B Notes A (5) B (4) Notes (0) View Transaction Table Item: A Item: B Item: Notes Order Form Order Type: ������ Price: 195.00 � ���� �� � Market: A ������������ Price: $195.00 ����������� Units: 1 �������������� ����� Total Value: $195.00 Messaging Messaging Received Messages mc : Markets will be called in 10 minutes mc : Markets will be called in 9 minutes mc : Markets will be called in 8 minutes mc : Markets will be called in 7 minutes Message History 11 ��������������� ��������������� ������������������������������ ����������������������������� CAPM PREDICTIONS • The expected returns of each of securities A and B are positively related to the betas of the securities • The market portfolio is mean-variance efficient. • The final portfolio of each individual should have risky securities in the same proportion as in the market portfolio. 12 6&

  7. 7/26/12& Prices • Prices of the two securities are virtually the same. • Expected payoff of A is higher, i.e., expected return of A is higher. • Precisely because Beta(A)>Beta(B) 13 Efficiency of market portfolio • Compute Sharpe ratio of market portfolio with each transaction. • Take the difference between Sharpe of the market and the highest Sharpe ratio. • CAPM predicts this difference should be 0. 14 7&

  8. 7/26/12& INDIVIDUAL Behavior • For each of the 8 periods plot each 1 subject’s final holdings of asset A. 0.8 • Red dot indicates the 0.6 proportion proportion of A in the market portfolio. 0.4 • Individuals are all 0.2 over the place. Allocations do not 0 improve with time. 1 2 3 4 5 6 7 8 period 15 2. Risk prediction in the human brain • How does the human brain process risk? • Specifically, what are the computational algorithms that the brain uses? • How general are these (algorithms)? Any potential flaws (that would explain cognitive biases)? 16 8&

  9. 7/26/12& On cognitive biases • The human brain is a remarkably adapted computational device • But it does not do everything right • Example • Knowing whether you are level when flying through clouds • Pilots need instrument rating to get permission to fly through clouds) • Important: there is NO WAY to teach the brain to do this right! You need instruments 17 a simple gamble • With each card, update: • EXPECTED REWARD • RISK 18 9&

  10. 7/26/12& Risk and reward • Expected reward increases linearly • Risk (reward variance) is quadratic 19 Functional magnetic resonance imaging (fMRI) • Subjects play game repeatedly while brain is being scanned in fMRI • Measures brain activation indirectly (through blood flow) 20 10&

  11. 7/26/12& Neural activation correlating with risk • fMRI signal per level of reward probability; averaged across subjects • Activation AFTER seeing first card 21 Anterior insula • Known to “convert” emotions into “feelings” • (Ability to sense heartbeat ~ size) • Related to self-awareness • Activates especially in reaction to disgust, pain Cool mathematics in a quintessentially emotional • Involved in empathy part of the human brain! 22 11&

  12. 7/26/12& Neural activation correlating with risk prediction error • Traditional reward prediction error: reward minus expected reward • Risk = expectation of size of reward prediction error • Risk prediction error 23 fMRI evidence Risk prediction error after first and second card. No risk, and hence, no risk prediction error if first card is 1 or 10 24 12&

  13. 7/26/12& Effectively, brain signals whether there are outliers • Is size of realized reward prediction error bigger than expected? • If so, things may have changed… • Regime shifts • This, however, is not always the right interpretation! Black swans are not unusual… • Does the brain distinguish? 25 Pupil dilation • Pupil dilation • … correlates with risk prediction error • There is a link with phasic changes in norepinephrine levels • Could study this pharmacologically (propranolol?) 26 13&

  14. 7/26/12& Final remarks • We are bringing finance to the laboratory, in order to distinguish what is intrinsically wrong from what is wrong because we don’t know the parameters. • We have seen some fundamentals concepts in asset pricing theory at work, like CAPM. What we learned: the system has its own laws, different from the individuals • We started exploring things where the theory is not quite right • We have seen excessive volatility, but it does not affect allocations (welfare)! • We are beginning to understand how markets equilibrate • We have started to explore the neurobiological foundations of cognitive biases – the brain uses specific algorithms to track the environment, and these may not always be well adapted to financial markets • Implementation of the algorithms often engages emotions – emotions are part of “reasoned” decision making 27 14&

Recommend


More recommend