Neural$Dynamics$ofDecision:Making$in$a$Financial$Trading… · 2014. 3. 31. · Subject-level...
1
Occipital pTemporal Parietal Central aTemporal Frontal Time Window (ms) –3.5 3.5 T Values 300 600 900 1200 1500 0 0.05 P Values 300 600 900 1200 1500 ERP Results: Choice Optimality Neural Dynamics of DecisionMaking in a Financial Trading Task Alison Harris 1 , Cary Frydman 2 & Tom Chang 2 1 Department of Psychology, Claremont McKenna College, Claremont, CA USA 2 Department of Finance and Business Economics, Marshall School of Business, University of Southern California, Los Angeles, CA USA ❖ Disposition effect ❖ Sell winning stocks more often than losing stocks ❖ Deviation from optimal financial decision-making ❖ Realization Utility theory ❖ Pleasure from sale relative to purchase cost (capital gain) ❖ It hurts to sell at a loss, but “locking in” a gain is satisfying ❖ When and how does financial decision-making occur in the brain? ❖ Value signals related to capital gain ❖ Neural correlates of optimal vs. suboptimal trading choice Introduction } Methods ❖ N = 60 ❖ Investing in stock market with stocks A, B, C ❖ Update period: Price change ❖ Action period: Buy or sell decision $80 B BUY? Action 2 DOWN $5 --> $80 B Don’t own Update 2 $75 A Cost: $100 SELL? Action 1 Update 1 UP $15 --> $30 C Purchased: $100 Trial str ructure Update 2 s Prep (Fixation) 1-2 s Action RT (max 3 s) ITI (Fixation) 3 * RT ❖ Procedure ❖ 128-channel EEG ❖ Could only hold 0 or 1 units of each stock ❖ Informed of stock market properties at start of experiment ❖ Good state: p(up) = 0.7, p(down) = 0.3 ❖ Bad state: p(up) = 0.3, p(down) = 0.7 ❖ 20% chance of changing from good to bad state or vice versa ❖ Payoff at end of experiment based on stock holdings and sales Predictions ❖ Participants will exhibit behavioral disposition effect (DE) ❖ Capital gain (CG) at sell decision correlates with neural value signal ❖ Ventromedial prefrontal cortex (vmPFC) ❖ From ~400 ms after stimulus onset ❖ Neural sensitivity to CG associated with selling “winners” ❖ Optimal choice requires overcoming realization utility bias ❖ Analogous to regulating behavioral/cognitive conflict ❖ Anterior cingulate cortex (ACC) ERP Results: Capital Gain Behavioral Results ❖ Average DE = 0.07 significantly greater than zero (p = 0.04) ❖ Suboptimal behavior compared to “rational” Bayesian agent -0.7 -0.5 -0.3 -0.1 0.1 0.3 0.5 0.7 Disposition Effect Subject “Rational” DE 0 1250 2500 3750 5000 Held Gains Sold Gains Held Losses Sold Losses Number of Responses Optimal Suboptimal Neutral ❖ EEG data time-locked to Action period onset ❖ Subject-level linear regression: Capital Gain: y sensor,time = β 0 + β 1 CapitalGain + β 2 BayesianPosterior + ε 100 ms 100 ms + 0.5 μV – Lateral Central Optimal Suboptimal Hold Sell -0.4 μV -0.8 μV 0 μV 1.2 μV Gain Loss * ❖ EEG data time-locked to Action period onset ❖ Paired t test on optimal vs. suboptimal choice ❖ Optimal: hold if b i (good) > 0.5 , sell if b i (good) < 0.5 150 ms Frontal 100 ms + 1 μV – 1.8 μV 2.1 μV Mean Amplitude <25% Gain 25-50% Gain 50-75% Gain >75% Gain –3 3 T Values 150 to 200 ms post-stimulus 400 to 650 ms post-stimulus 450 ms Central 100 ms + 0.5 μV – 0.9 μV 1.3 μV Mean Amplitude 350 to 450 ms post-stimulus -0 μV -0.8 μV 0 μV 1.2 μV Gain Loss ❖ Disposition effect exists despite being financially suboptimal ❖ Capital gain at sell decision correlates with ERP value signal ❖ Emerges 400-650 ms after stimulus onset ❖ Localized to vmPFC ❖ Neural CG signal correlates with propensity to sell winners ❖ Optimal choice requires overcoming realization utility bias ❖ Neural signals as early as 100-150 ms after stimulus onset ❖ Localized to ACC ➡ ERP provides insight into time course of disposition effect ➡ Supports role of neural value signals in realization utility bias Conclusions Occipital pTemporal Parietal Central aTemporal Frontal Time Window (ms) –3.5 4 T Values 300 600 900 1200 1500 0 0.05 P Values 300 600 900 1200 1500 100 to 150 ms post-stimulus Correlating ERP with Behavior 0 0.2 0.4 0.6 0.8 1 -1 -0.5 0 0.5 1 Propensity to Sell Winners Capital Gain β 400-650 ms r = 0.28 p = 0.03 0 0.2 0.4 0.6 0.8 1 -1 -0.5 0 0.5 1 Propensity to Sell Losers Capital Gain β 400-650 ms r = 0.05 p = 0.45 ERP Source Reconstruction ❖ Distributed source reconstruction in SPM8 (group inversion) Capital Gain, 400-650 ms post-stimulus fMRI ❖ Linear ordering of CG quartiles ❖ Localized to vmPFC ❖ Consistent with fMRI (Frydman et al., 2014) F = 150 y = 39 Optimal vs. Suboptimal, 100-150 ms post-stimulus ❖ Optimal > Suboptimal ❖ Left dorsal ACC ❖ Bilateral precentral gyrus ❖ Left anterior insula ❖ Suboptimal > Optimal ❖ Genual ACC ❖ Right anterior insula ❖ Bilateral parietal lobe 5 6 –6 –5 T Values vmPFC 100 ms + 0.5 μV – Lateral Central 350 ms Mean Amplitude Mean Amplitude
Transcript of Neural$Dynamics$ofDecision:Making$in$a$Financial$Trading… · 2014. 3. 31. · Subject-level...
Occipital
pTemporal
Parietal
Central
aTemporal
Frontal
Time Window (ms)
–3.5
3.5
T Values
300 600 900 1200 1500
0
0.05
P Values
300 600 900 1200 1500
ERP Results: Choice Optimality
Neural Dynamics of Decision-‐Making in a Financial Trading TaskAlison Harris1, Cary Frydman2 & Tom Chang2
1 Department of Psychology, Claremont McKenna College, Claremont, CA USA2 Department of Finance and Business Economics, Marshall School of Business, University of Southern California, Los Angeles, CA USA
❖ Disposition effect❖ Sell winning stocks more often than losing stocks❖ Deviation from optimal financial decision-making
❖ Realization Utility theory❖ Pleasure from sale relative to purchase cost (capital gain)❖ It hurts to sell at a loss, but “locking in” a gain is satisfying
❖ When and how does financial decision-making occur in the brain?❖ Value signals related to capital gain❖ Neural correlates of optimal vs. suboptimal trading choice
Introduction
}
Methods❖ N = 60❖ Investing in stock market with stocks A, B, C
❖ Update period: Price change❖ Action period: Buy or sell decision
$80B
BUY?
Action 2
DOWN $5 --> $80B
Don’t own
Update 2
$75A
Cost: $100 SELL?
Action 1
Update 1
UP $15 --> $30C
Purchased: $100
Trial structureTrial structureTrial structureTrial structure
Update2 s
Prep (Fixation)1-2 s
ActionRT (max 3 s)
ITI (Fixation)3 * RT
❖ Procedure❖ 128-channel EEG❖ Could only hold 0 or 1 units of each stock❖ Informed of stock market properties at start of experiment
❖ Good state: p(up) = 0.7, p(down) = 0.3❖ Bad state: p(up) = 0.3, p(down) = 0.7❖ 20% chance of changing from good to bad state or vice versa
❖ Payoff at end of experiment based on stock holdings and sales
Predictions❖ Participants will exhibit behavioral disposition effect (DE)❖ Capital gain (CG) at sell decision correlates with neural value
signal❖ Ventromedial prefrontal cortex (vmPFC)❖ From ~400 ms after stimulus onset❖ Neural sensitivity to CG associated with selling “winners”
❖ Optimal choice requires overcoming realization utility bias❖ Analogous to regulating behavioral/cognitive conflict❖ Anterior cingulate cortex (ACC)
ERP Results: Capital Gain
Behavioral Results❖ Average DE = 0.07 significantly greater than zero (p = 0.04)❖ Suboptimal behavior compared to “rational” Bayesian agent
-0.7
-0.5
-0.3
-0.1
0.1
0.3
0.5
0.7
Dis
po
siti
on
Eff
ect
Subject
“Rational” DE
0
1250
2500
3750
5000
Held Gains Sold Gains Held Losses Sold Losses
Num
ber
of R
esp
ons
es
OptimalSuboptimalNeutral
❖ EEG data time-locked to Action period onset❖ Subject-level linear regression:
Capital Gain: ysensor,time = β0 + β1CapitalGain + β2BayesianPosterior + ε
99 ms
100 ms
100 ms
+0.5 µV
–
Lateral
Central
OptimalSuboptimal
HoldSell
-0.4 µV
-0.8 µV
0 µV
1.2 µV
Gain Loss
*
❖ EEG data time-locked to Action period onset❖ Paired t test on optimal vs. suboptimal choice
❖ Optimal: hold if bi(good) > 0.5 , sell if bi(good) < 0.5
150 ms
150 msFrontal
100 ms
+1 µV
–1.8 µV
2.1 µVMean Amplitude
<25% Gain 25-50% Gain 50-75% Gain >75% Gain
–3
3
T V
alue
s
150 to 200 ms post-stimulus
400 to 650 ms post-stimulus
450 ms
450 ms
Central
100 ms
+0.5 µV
–0.9 µV
1.3 µVMean Amplitude
350 to 450 ms post-stimulus
-0 µV
-0.8 µV
0 µV
1.2 µV
Gain Loss
❖ Disposition effect exists despite being financially suboptimal❖ Capital gain at sell decision correlates with ERP value signal
❖ Emerges 400-650 ms after stimulus onset❖ Localized to vmPFC❖ Neural CG signal correlates with propensity to sell winners
❖ Optimal choice requires overcoming realization utility bias❖ Neural signals as early as 100-150 ms after stimulus onset❖ Localized to ACC
➡ ERP provides insight into time course of disposition effect➡ Supports role of neural value signals in realization utility bias
Conclusions
Occipital
pTemporal
Parietal
Central
aTemporal
Frontal
Time Window (ms)
–3.5
4
T Values
300 600 900 1200 1500
0
0.05
P Values
300 600 900 1200 1500
100 to 150 ms post-stimulus
Correlating ERP with Behavior
0
0.2
0.4
0.6
0.8
1
-1 -0.5 0 0.5 1
Pro
pen
sity
to S
ell W
inne
rs
Capital Gain β 400-650 ms
r = 0.28p = 0.03
0
0.2
0.4
0.6
0.8
1
-1 -0.5 0 0.5 1
Pro
pen
sity
to S
ell L
ose
rs
Capital Gain β 400-650 ms
r = 0.05p = 0.45
ERP Source Reconstruction❖ Distributed source reconstruction in SPM8 (group inversion)
Capital Gain, 400-650 ms post-stimulus
fMRI ❖ Linear ordering of CG quartiles
❖ Localized to vmPFC❖ Consistent with fMRI
(Frydman et al., 2014)F = 150
y = 39
Optimal vs. Suboptimal, 100-150 ms post-stimulus
❖ Optimal > Suboptimal❖ Left dorsal ACC❖ Bilateral precentral gyrus❖ Left anterior insula
❖ Suboptimal > Optimal❖ Genual ACC❖ Right anterior insula❖ Bilateral parietal lobe
5 6–6 –5
T Values
vmPFC
100 ms
+0.5 µV
–
Lateral
Central
Latency 350 ms from TVals
−3.1
−1.6
0
1.6
3.1
350 ms
Mean Amplitude
Mean Amplitude
![[width=3.3cm]images/LogoMOA.jpg .5cm Regressionabifet/523/Regression-Slides.pdf · I Regression Gain = SD(before Split) SD(after split) StandardDeviation (SD) = qX (y yi)2=N. Numeric](https://static.fdocuments.us/doc/165x107/5f06e0f57e708231d41a3040/width33cmimages-5cm-regression-abifet523regression-slidespdf-i-regression.jpg)
