1 Decision Analysis Scott Matthews /
and Administrative Comments Group Project 1 Back - Average 91% How graded? High level thoughts - good on NPV Some missed big picture - NPV? HW 3 due next Wednesday
and Commentary It is trivial to do “economics math” when demand curves, preferences, etc. are known. Without this information we have big problems. Unfortunately, most of the ‘hard problems’ out there have unknown demand functions. We need advanced methods to find demand
and Estimating Linear Demand Functions zAs above, sometimes we don’t know demand zFocus on demand (care more about CS) but can use similar methods to estimate costs (supply) zOrdinary least squares regression used yminimize the sum of squared deviations between estimated line and p,q observations: p = a + bq + e yStandard algorithms to compute parameter estimates - spreadsheets, Minitab, S, etc. yEstimates of uncertainty of estimates are obtained (based upon assumption of identically normally distributed error terms). zCan have multiple linear terms
and Also - Log-linear Function zq = a(p) b (hh) c ….. zConditions: a positive, b negative, c positive,... zIf q = a(p) b : Elasticity interesting = (dq/dp)*(p/q) = abp (b-1) *(p/q) = b*(ap b /ap b ) = b. yConstant elasticity at all points. zEasiest way to estimate: linearize and use ordinary least squares regression (see Chap 12) yE.g., ln q = ln a + b ln(p) + c ln(hh)..
and Log-linear Function q = a*p b and taking log of each side gives: ln q = ln a + b ln p which can be re-written as q’ = a’ + b p’, linear in the parameters and amenable to OLS regression. Alternative is maximum likelihood - select parameters to max. chance of seeing obs.
and Maglev Log-Linear Function q = a*p b - From above, b = -0.3, so if p = 1.2 and q = 20,000; so 20,000 = a*(1.2) -0.3 ; a = 21,124. If p becomes 1.0 then q = 21,124*(1) -0.3 = 21,124. Linear model - 21,000 Remaining revenue, TWtP values similar but NOT EQUAL.
and Structuring Decisions All about the objectives (what you want to achieve) Decision context: setting for the decision Decision: choice between options (there is always an option, including status quo) Waiting for more information also an option Uncertainty: as we’ve seen, always exists Outcomes: possible results of uncertain events Many uncertain events lead to complexity
and Structuring Decisions (2) Can use: Fundamental objective hierarchy. Influence diagrams. Decision Trees Risk Profiles
and Fundamental Objectives Hierarchy Increase Lifetime Earnings Increase Current Salary Find New JobGet a Raise Update ResumeNetworkDo a Better Job Marry RichGo to School UndergradGrad School
and Influence Diagram/Decision Trees Probably cause confusion. If one confuses you, do the other. Important parts: Decisions Chance Events Consequence/payoff Calculation/constant
and Influence Diagram Lifetime Earnings Work High Salary Get a Raise Find a Better Job Marry Rich Go to School Undergrad Grad School
and Other Notes Chance node branches need to be mutually exclusive/exhaustive Only one can happen, all covered “One and only one can occur” Timing of decisions along the way influences how trees are drawn (left to right) As with NPV, sensitivity analysis, etc, should be able to do these by hand before resorting to software tools.
and Solving Decision Trees We read/write them left to right, but “solve” them right to left. Because we need to know expected values of options before choosing. Calculate values for chance nodes Picking best option at decision nodes We typically make trees with “expected value” or NPV or profit as our consequence Thus, as with BCA, we choose highest value.
and Texaco vs. Pennzoil Counteroffer $5 Billion Texaco Counteroffer $3 Billion Refuse Settlement Amount ($ Billion) Accept $2 Billion 2 Texaco Accept $5 Billion 5 Texaco Refuses Counteroffer Final Court Decision Final Court Decision Accept $3 Billion 3 (0.17) (0.5) (0.33) (0.2) (0.5) (0.3) (0.2) (0.5) (0.3)
and To Solve the Tree Solve from right to left: At chance node multiply monetary value to probability and add them. At choice node choose highest value. EMV for Simple Texaco vs. Pennzoil Tree: $4.63 Billion
and Risk Profiles Risk profile shows a distribution of possible payoffs associated with particular strategies. A strategy is what you plan to do going in to the decision. Holds your plans constant, allows chances to occur Only eliminate things YOU wouldn’t do, not things “they” might not do. Its not just finding the NPV of a branch.
and Risk Profiles (cont.) Let’s think about the “subset” of the Texaco decision tree where we are only curious about the uncertainty/risk profile associated with various strategies to consider. These represent the riskiness of each option There are only 3 “decision strategies” in the base Texaco case: Accept the $2 billion offer (topmost branch of 1st dec. node) Counteroffer $5 Billion, but plan to refuse counteroffer (lower branch of 1st node, upper branch of second) Counteroffer $5B, but plan to accept counteroffer (lower branch of both decision nodes)
and Texaco vs. Penzoil, Again Risk profile for “Accept $2 Billion” is obvious - get $2B with 100% chance.
and Risk Profile: Counteroffer $5, accept $3 billion Below is just the part of original tree to consider when calculating the risk profile: Counteroffer $5 Billion Texaco Counteroffer $3 Billion Texaco Accept $5 Billion 5 Texaco Refuses Counteroffer Final Court Decision Accept $3 Billion 3 (0.17) (0.5) (0.33) (0.2) (0.5) (0.3)
and Texaco vs. Pennzoil, continued
and Cumulative Risk Profiles Graphs of cumulative distributions Percent chance that “payoff is less than x”
and Dominance To pick between strategies, it is useful to have rules by which to eliminate options Let’s construct an example - assume minimum “court award” expected is $2.5B (instead of $0). Now there are no “zero endpoints” in the decision tree.
and Dominance Example CRP below for 2 strategies shows “Accept $2 Billion” is dominated by the other.
and Next Class Value of Information. Facility Case Due