Weeks 15-16: Conglomerates, Lemmings, Founders’ Dilemmas and Tax Lotteries

Stanford GSB Sloan Study Notes, Week 5-6 (15-16), Autumn quarter

This post consolidates my notes from two weeks instead of a normal one, yet will be a bit more concise than usual too, for a few reasons: I was down with flu for several days and had to miss a few classes and then the midterm exams in Financial Accounting and Organizational Behaviour changed the normal scheduling.

Also, the first session of the latest addition in our core timetable, STRAMGT 259: Generative Leadership by Dan Klein yesterday was too… experiential to take any notes, really. Basically, we did three hours of improv theatre. It was a lot of fun, but instead of getting into the theory here – get the book: Improv Wisdom: Don’t Prepare, Just Show Up by Patricia Madson. And say “yes” more to whatever life throws at you, go with the flow and see what happens.

For additional entertainment, here is an experiment shared by my classmate Marc who is lucky to take a Behavioral & Experimental Economics class by freshly Nobel-prized Al Roth: primatologist Frans de Waal showing how even monkeys reject unequal pay (see especially from 2nd minute).

And now on to the regular programming. Covered in this issue:

Why people suck at predicting when they finish a task
How overdiversification, and especially uncontrolled aquisitions lead to dysfunctional conglomerates
Lemmings following lemmings, but not sheep
Predicting future divorces
Research from surveying 10,000 founders that quantifies the impact of common “gut decisions” like picking investors or sharing stock between co-founders
Guest speakers explaining how they’ve used creative incentive schemes to get more out of porn site classification crowdsourcing and VAT payments in China
The impact of investment lags on IP value creation in startups and established companies

GSBGEN378 – Decisions About the Future (Hardisty)

Planning fallacy – tendency for people to underestimate how long it will take for them to complete a task
- Article: Exploring the “Planning Fallacy”: Why People Underestimate Their Task Completion Times
- Actor-observer differences: more realistic when predicting how others perform, rather than you
- Even if there is strong optimistic bias between bestguess/optimistic/pessimistic predictions, even the worst-off optimistic guesses are not devoid of information (highly correlated with actual results still)
- Likewise, subjects who self-rate higher confidence, will actually perform better
- When pondering out loud, most of focus on future planning (71%) or future problems (3%) in building prediction. Only 7% references to past problems/successes.
  - Yet, more realistic estimates when a) recalling past experiences and b) relating those to task at hand.
Save More Tomorrow – using behavioural economics to increase employee saving
- Convince people to commit ahead to increasing their saving rate with every next salary increase
- Null hypothesis difference:
  - Standard economic model: workers have no interest in joining, because they have already chosen their optimal savings rate.
  - Behavioral economic model: workers will find this attractive and will join.
- Prescriptions often have a second-best quality – e.g, simplest advice might not be the best (too short therm, “quick-hack”)
- Hyperbolic agents procrastinate because they (wrongly) assume whatever they will do later is less important than first step. And procrastination produces inertia.
  - An investment study showing median number of changes in pension plan asset allocation over lifetime was zero. (More than half participants retire with same asset allocation they started saving with!)
  - In a mixed stock/bond portfolio, stocks appreciate much more over lifetime which throws allocation out of balance.
Libertarian paternalism: philosophy advocating design of institutions that help people make better decisions, but do not impinge on their freedom to choose (ex: opt out)
Tobias Preis: GDP vs future orientation correlation shown based on Google searches for past and future year numbers from different countries
- Future Orientation Index: raw data PDF
For top athletes, first few years of training focuses on identifying appropriate arousal (stress) levels
- then next years on timing peak performance using the right levels
- Ex: 10 days/year max, 2-3 months/year @ top 2% performance
Halo effect
- Same teacher ranked more handsome based on communication style
- can reduce framing/bias with explicit callouts (“I bet eating made you sleepy” in the beginning of afternoon meeting -> the other party doesn’t think they are sleepy because you are boring)
Visualization proven to increase people’s support to causes, ex: climate change
- Pictures higher impact than graphed data
- some causes are “lucky” to have powerful visuals more naturally available
- compare to Singularity Institute & AI research, NASA pitching for funding of a never-before mission

STRAMGT279 – Global Strategic Management (Roberts)

“Judo economics” – use the size of an incumbent as a weapon
- costly for a giant to swat a fly, so it lets the fly live (until not too bothersome)
Vertical integration to get returns from operational improvements
- Presumes long-term
- Hard to replicate contractually between independent parties (ex: consulting contract to share benefits of improvements)
Conglomerates are more alive in emerging markets because…
- capital markets less effective to diversity otherwise
- management skills in undersupply
- family ownership -> personal wealth diversification through M&A
Japanese keiretsu provided conglomerate-like coordination without ownership
- common bank, trading companies, regular cadence of meetings between chairmans…
Debt is hard and unforgiving, equity is soft.
- not enough cash for dividends? tough luck…
- not enough cash to repay your debt? default…
M&A is often motivated by CEO / executive interests, rather than shareholders’
- prestige, equity plans, total pay correlated strongly to company size
- integration success or failure harder to measure, longer term
Most M&A positive return for sellers, lawyers, bankers and negative for buyers.
- Winner’s curse: bidder who is most off when estimating value is most likely to win.
M&A spillovers test: _which specific underused assets_will the synergies come from post merger?
- management? knowledge? technology? sales force?
Diversification discount on conglomerate valuation
- tend to keep financing weak and underinvest in progressing BU-s
- due to management complexity, corporate politics, innovators dilemma…
Illinois had quite recently a law that only allowed banks with maximum of one branch!
Xerox
- using executive assistant roles as mentorships
  - “talent development on steroids. 1-2 year assignments to go deep and broad learning how the org operates through the lens of a senior executive”
- family metaphore for culture (Burns): “When you’re inside the family, you don’t have to be as nice as outside. Let’s stay civil and kind, but we have to be frank. And the reason why we can be frank is that we’re in the same family.”

OB278 – Organizational Behaviour (Flynn)

Psychology of blame
- Situational change from A (all was good) to B (now bad)
- Blame subconsciously assigned to what has changed – despite the change being relevant or now
- E.g. if there is a new person in the project, easily assumed to be to blame
Conflict resolution splits into
- Task goals – short term, fix the problem
- Relationship goals – long term, maintain the relationships in the process
- 2 often in conflict & one shall dominate
  - The more interactive/intimate the communication, the more likely relationship goals prevail (face-to-face chat VS email)
Agreeing to a protocol increases trust to the contract, not the other party
- Why legal and ethical/moral solves are often so different?
Reacting to problems interpreted as “has to”, while anticipating (being proactive) as “wants to” (by clients, team, bosses)
- drop present-focus (from task -> relationship goal)
- find out _who_the other party is ASAP
  - needs (to be liked, have status, achieve X…)
  - priors (CV, jobs, but more specifically successes…)
  - views (of self, you, world)
- Identity markers around people’s physical location (clothing, workspace design & decorative items, …)
Book: Divorce in Psychosocial Perspective: Theory and Research by Joseph Guttman
- Experimental research into what causes long-term relationship breakdowns and how to predict
- Spoiler: highest correlating predictors are asynchronous signals during conflicts between parties (eye rolls, smirking)
Height <-> authority link proven both ways
- experiment: guess the height of a speaker introduced with different titles to different audiences
- “grad student” perceived to be 2″ shorter than the same person as “tenured professor”
- talk show hosts seated 6-8″ higher than guests (and thus the hiding behind a desk)
Canned laughter in TV sitcoms
- less about prompting the “right time to laugh”
- more to compensate for increasing trend to watch TV alone (and thus not laugh)
- TV ratings proven to increase in the singles segment, not across the board
“Salting” the tip jar
- defines the norm, not just “tips or not”, but also “how much others consider appropriate”
Milgram’s sidewalk experiment:
- a single person staring in the sky in NYC makes people flee (madman?), 5 people doing that simultaneously draws a large crowd who also start staring up
Lemmings effect is amplified by narrowing the category of reference
- relevance doesn’t matter (again!) – “people with the same birthday as you”
- “people staying in this room” more effective than “… in this hotel”
“Sheep follow sheep, not lemmings”
Same data can be framed based on which subset of the group you need the lemming to join, e.g for stealing:
- deviant framing: “less than 2% of our visitors steal” -> eliminated stealing
- normative frame: “more than 1000 visitors steal” -> 3X more stealing
Both popularity and _momentum_matter to create following
- manufactured social proof: author buying their own book to hit the charts (after which natural sales follow)
- Yahoo! experiments showing download counts to free MP3-s had strong cascading influence after some song got just slightly ahead
George Lakoff – UC Berkeley research on language use in persuasion
Diffusion of responsibility in groups
- in emergency shouting “HELP!!!” is less effective than “YOU, with the glasses, call 911!”
- a group reminder is less effective than personal (“is it just me?”)

**MS&E 472 – **Entrepreneurial Thought Leaders

Guest: Noam Wasserman, The Founder’s Dilemmas

Book: The Founder’s Dilemmas: Anticipating and Avoiding the Pitfalls That Can Sink a Startup (Kauffman Foundation Series on Innovation and Entrepreneurship)
- “Drawing on a decade of research, Noam Wasserman reveals the common pitfalls founders face and how to avoid them. He looks at whether it is a good idea to cofound with friends or relatives, how and when to split the equity within the founding team, and how to recognize when a successful founder-CEO should exit or be fired. Wasserman explains how to anticipate, avoid, or recover from disastrous mistakes that can splinter a founding team, strip founders of control, and leave founders without a financial payoff for their hard work and innovative ideas. He highlights the need at each step to strike a careful balance between controlling the startup and attracting the best resources to grow it, and demonstrates why the easy short-term choice is often the most perilous in the long term.
- The Founder’s Dilemmas draws on the inside stories of founders like Evan Williams of Twitter and Tim Westergren of Pandora, while mining quantitative data on almost ten thousand founders.”
Startup failure reasons:
- 35% related to product-market-fit (product mgmt, functional mgmt, market problems)
- 65% people problems (interpersonal tensions, between founders and/or with hired employees)
Data on 10,000 founders
Core founder dilemmas
- When to found? – core founder
  - Kaufmann Foundation’s sketchbook for pre-founding decisions
  - Only 16% of start-ups are solo-founded
- Building the team (co-founders)
  - More than 50% of startups have co-founders who know socially (friends & family), but not professionally
    - Paradoxically, data shows despite of these social ties, these founder teams are more likely to break down
  - Founder teams that start with seeking decision making consensus, co-CEO setups, one-founder-one-vote struggle with adjusting to needed role divisions later
  - Relationships-roles-rewards tension: co-founding with your best friend on consensus model -> can’t go and say you want 60% of stock
  - 73% of startups split equity between founders in the first month on a static agreement
    - 33% equal split
    - 12% near-equal (1-10% diff)
    - 15% moderate gap (11-20%)
    - 19% large gap (21-40%)
    - 21% huge gap (>40%)
  - Can you really do a static contract? Is the strategy stabilised, business model set? Roles & skill requirements known?
    - Founder time commitment, personal uncertainties?
    - “just because you’re founding, doesn’t mean life events will stop happening to you”
  - Conditional contracts listing commitment (full- or part-time?), milestones as if-then decision trees, with buyout terms as the worst case scenario
- New-Venture hiring
- Beyond the team (investors, partners)
  - Board control as investment consideration
  - 50 months from founding less than 50% of founders are still in the business, in 100 months: 25% left
    - 73% of that are firings
  - Control vs financial tradeoff: 52% more value retained by founders who give up both CEO and board control
    - Do you want to be the King or Rich?
- Exit dilemmas

CS547: Human-Computer Interaction Seminar

Note: full archive of videos for this course available here.

Guest speaker: Panos Ipeirotis (NYU Stern School of Business), Crowdsourcing: Achieving Data Quality With Imperfect Humans – @ipeirotis

Blog: Behind Enemy Lines -> Computer Scientist in Business School
Problem: building complex targeting models _fast_ – plain linguistic analysis not enough
- Ex: plane ads next to articles near crast inappropriate; FDA banning drug ad placement next to swine flu coverage
- AdSafe “adult web site classifier”: asking people to categorise sites in four buckets (general audience .. porn)
  - Undergrad: 200 siters / hr, cost $15/h
  - Amazon Mechanical Turk: 2500 sites/hr, cost $12
  - BUT: quality dropping enormously and checking/fixing very expensive, manual
Expectation Maximization Estimation: establish “correct” labels by majority vote, estimate error rates for workers & weigh worker rates by quality
- Weigh down the answers of those who more often diverge from group
- Algorithm not novel, invented in 1979 with doctors & interns
- Challenge 1: spammers figure it out (0% * 85% + 100% * 15% = 15% spammer error rate is lower than good user: 85% * 20% + 85% + 20% = 20%)
- Challenge 2: human biases – what is offensive and what is not depends on role, culture, etc
Establish a quality score, a scalar measure that distinguishes spammers with unpredictable (random, constant) answers to any question from humans whose biases a predictable on data (e.g when answering A, spammer can mean anything, but a biased human means consistently B compared to peers)
Quality-sensitive payment
- Set goal: $1 for 90%
- For workers above: full pay
- For others: payment divided with redundancy needed to reach goal (3 workers for 90% => $0.33 pay)
Fair to reimburse for difference in past work quality as the understanding of quality improves
- Reduces churn!
“Google bias” – people submit URLs that you already know about being bad
- Beat the Machine: incentive to find URLs classifiers fail on – $1 if you beat the machine, $0.001 if you don’t
Difference between “unknown unknowns” & “known unknowns”
- e.g multilingual porn – once a new domain is submitted, it could be expected that derivatives in different languages also exist / are offensive
- Reduce rewards based on classifier confidence
Crowdsourcing workers behave like mice – cognitive reward system similar to food pellets:
- Bad answers introduce probabilistic nuisances: delays between tasks, timeouts
- Good answers introduce variety, change, visible improvements
- See: Misery module for Drupal
- Spammers witnessed to drop extra quickly and move on… and 15-20% spammers start submitting good work!
Book: Kahneman’s Thinking, Fast and Slow (again!)
Modern-day slot machine: email. People check it constantly because sometimes it pays out… and most of the time it does not.

Guest speaker: Balaji Prabhakar (EE/CS department, Stanford): Societal Networks

Societal networks
- meeting point of tech, policy & economics
- vital for society, ex: transportation, healthcare, energy, recycling, …
Less than 2% of US passenger miles in public transportation
- 48% SG, 94% HK
Turning small awards for good behaviour into a lottery
- $0.05 per each empty can returned to recycling VS 1/2000 chance of getting $100?
An experiment in some provinces of China (and Taiwan?), every VAT receipt is a lottery ticket
- 30M Yuan in rewards paid on 900M Yuan of tax collection! (98-02)
- Lottery Receipt Experiment in China (Junmin Wan, Fukuoka University)
Stanford Congestion and Parking Relief Incentives: https://stanfordcapri.org/
- Commute at off-peak hours, earn credits, redeem for rewards

CS207 – Software Economics (Wiederhold)

For Pareto optimal operations consider the investment lags
- Inhouse hiring vs contractors/outsourcing
- Private hosting vs cloud services
- Organic growth vs acquisitions
- New technology development vs more marketing for existing products
- The surrogate of value of speeding IP generation up is income, not cost
Counterintuitively: startups have a longer lag than established companies
- Latter can ramp up their resource spending faster
- NB! Presumes similar product complexity, start-ups looking faster is a function of scaled-down scope of MVP + later iteration?

—

For more posts on the Stanford GSB Sloan life – click here to search by tag “sloan”.