Subscription / Ad financing: with application to US magazines · 3 3,5 20032005200720092011 0,5 0,7 0,9 1,1 1,3 1,5 1,7 1,9 2,1 20032005200720092011 Outdoor & sports Men’s Interior

Subscription / Ad financing: with application to US magazines

(pricing in the US Magazine Industry: A Two-Sided Market Perspective)

Simon Anderson and

Matthew Ce Shi University of Virginia

CESF Naples: Media Economics conference 2014

Background (Handbook of Media Economics Ch.)

• Changing face of media financing business model with onset of www

• How has mag subscription/ad finance mix changed recently?

• Reconcile with theory

• Showcase different business models and 2-sided pricing effects

• Ad-only; subscription only; joint (mixed) model

• sub price U-shaped as function of ad nuisance

• Want to estimate key parameters

- Is there a U-shape (across different mags/genres )?

- How have parameters changed over last decade?

– progress so far

Outline [“lit”]

• THEORY of equilibrium business model

(extended AC monopoly, MH advertisers – SH readers;

stress ad-loving and 2SM properties [MA review])

• Document patterns and recent changes for US mags

[Chandra-Kaiser Ch. in HoME]

• Preliminary results for some parameters

[prescient Kaiser-Wright IJIO06 Hotelling German mags estimation; Lapo]

• Future intentions

Monopoly model – ad finance only (AC)

• Π(a) = aP with P as Price PER AD

• Now write in readers, N(a)

• So Π(a) = ap(a)N(a) = R(a) N(a)

• foc

N

N

R

R ''

AC model: Linearized version

• Suppose that p(a)=1-a

• Then R(a)=a(1-a) and R’(a)=(1-2a)

• Suppose that N=1-f with f = s+γa

• For s=0, we have foc :

• Which becomes:

(1-2a)/a(1-a) = γ/(1- γa)

Plotted below …

N

N

R

R ''

Pure subscription finance

• If γ too large, we get no ads

• Then a simple monopoly (or oligopoly) pricing problem

• Linear demand; price at half the demand intercept (s=1/2)

Mixed finance

• Revenue per viewer is s+R(a);

• For any nuisance level, f= s+γa

the optimal split has γ=R’(a*)

So if γ>R’(0) there are no ads

-- Otherwise, for s>0 we have a* and

Π= (s+R(a*)) N(s+γa*)

Like a negative marginal cost carry-over!

(plus a fixed fee to viewers)

Mixed finance: Linearized version

• optimal split has γ=R’(a*) so a*=(1- γ)/2

So if γ>1 there are no ads

R(a*)=(1-a*)a*= (1- γ2)/2

-- Otherwise, for s>0 we have a* and

Π= (s+R(a*)) N(s+γa*) = (s+ (1- γ2)/2 ) (1-(s+γa*))

So that s satisfies: (1-(s+γa*)) = (s+R(a*))

Or s (>0) has s = (1-R(a*)-γa*)/2

Otherwise, s=0 and we have the ad-only regime

Pictures for this regime::::::->->->

ads: p=1-a; readers N=1-a; function of γ

ads: p=1-a; readers N=1-a; function of γ

Fees and nuisance

• Ad level falling throughout

• Subscription rising away from middle

• For γ <0 we have people like ads

• Don’t want too many because then hurts ad revenue

• For γ=-1 all viewers on-board

• Equivalently, for γ<-1 the viewer full price is negative and the ad level is above 1 (price below MC!)

• Viewers love ads so platform takes a loss on the ads and gets revenue from viewers / readers

• To profit and subs/ad finance ratio:

Stronger ad market

• With the earlier parameters there is no ad-only regime: the ad demand isn’t strong enough to supplant subscriptions completely

• Mags, newspapers, pay TV …

• With stronger ad demand, we get the ad-finance regime in the middle range

• So consider p=3-a

“large” ad mkt p=3-a; readers N=1-a; fn of γ

Large ads: p=3-a; readers N=1-a; fn of γ

Data • Documentation

- 83 US magazines for 11 years; representative of major genres; mags with largest reader base

- All price info is real (average) paid price instead of listed prices

- We see, on average, subs price decreasing, subs level constant/slightly increasing; ad price increasing, ad pages constant except a dip in 2007

• Estimation

- Use rich county-level sales information to estimate reader demand; allows “full” specification with mag dummies and time dummies

- Empirical equations come from theory model with closed-form solutions. We use data to “calibrate” the model.

Data Variable(s) # of obs. Mean Std. Dev.

MSA-level subscription sales (03-12)

246634 3667 10619.62 (min: 0, max: 876025)

Subscription level (nationwide, 02-12)

83 magazines*11 years

1234029 1190575

Subscription price (average paid, 02-12)

83*11 $22.41 $18.27

Ad volume (total ad pages, 02-12)

83*11 1142 577

Ad price per-page per-reader (02-12)

83*11 $ .14 $.09

Other characteristics (# of issues, content pages, avg. subscriber demo, 03-12)

83*10 - -

MSA demographics (ACS, 03-12)

- - -

Data Preview: Subscription Market

.51

1.5

22

.53

Su

bscri

ptio

n P

rice (

$, d

efla

ted)

2002 2004 2006 2008 2010 2012Year

Mean subscription price 2002: $2.20

Mean and Dispersion of Subscription Price: 2002-2012

0

100

00

00

200

00

00

300

00

00

# o

f sub

scrib

ers

2002 2004 2006 2008 2010 2012Year

Mean subscription 2002: 1,217,278

Mean and Dispersion of Subscription Level: 2002-2012

Magazine markets: subscription prices

0,8

1

1,2

1,4

1,6

1,8

2

2003 2005 2007 2009 2011

0,8

1,3

1,8

2,3

2,8

3,3

2003 2005 2007 2009 2011

0,5

1

1,5

2

2,5

2003 2005 2007 2009 2011

0

0,5

1

1,5

2

2,5

3

20032005200720092011

Women health Sci. & tech. Entertainment Parenting

0,5

1

1,5

2

2,5

2003 2005 2007 2009 2011

0,5

0,7

0,9

1,1

1,3

1,5

1,7

1,9

2,1

2,3

2003 2005 2007 2009 2011

1

1,5

2

2,5

3

3,5

2003 2005 2007 2009 2011

0,5

0,7

0,9

1,1

1,3

1,5

1,7

1,9

2,1

20032005200720092011

Outdoor & sports Men’s Interior design General interest

1

1,5

2

2,5

3

3,5

2003 2005 2007 2009 2011

0,5

1

1,5

2

2,5

3

2003 2005 2007 2009 2011

0,8

1

1,2

1,4

1,6

1,8

2

2,2

2,4

2003 2005 2007 2009 2011

0,8

0,9

1

1,1

1,2

1,3

1,4

1,5

1,6

20032005200720092011

Food Women fashion Business Automobile

Magazine markets: Subscription level

Women health Sci. & tech. Entertainment Parenting

Outdoor & sports Men’s Interior design General interest

Food Women fashion Business Automobile

5,00E+05

1,00E+06

1,50E+06

2,00E+06

2,50E+06

3,00E+06

3,00E+05

5,00E+05

7,00E+05

9,00E+05

1,10E+06

1,30E+06

1,50E+06

1,00E+04

5,10E+05

1,01E+06

1,51E+06

1,50E+06

2,00E+06

2,50E+06

3,00E+06

3,50E+06

4,00E+06

4,50E+06

5,00E+06

5,50E+06

6,00E+06

100000

600000

1100000

1600000

2100000

2600000

3100000

3600000

100000

600000

1100000

1600000

2100000

2600000

3100000

300000

400000

500000

600000

700000

800000

900000

1000000

100000

2100000

4100000

6100000

8100000

10100000

12100000

8,00E+05

1,00E+06

1,20E+06

1,40E+06

1,60E+06

1,80E+06

300000

500000

700000

900000

1100000

1300000

1500000

1700000

1900000

2100000

2300000

2003 2005 2007 2009 2011

400000

600000

800000

1000000

1200000

1400000

1600000

1800000

2000000

400000

600000

800000

1000000

1200000

1400000

Data Preview: Advertising Market

500

100

01

50

02

00

02

50

0

Ad

Leve

l (p

age

#)

2002 2004 2006 2008 2010 2012Year

Mean ad level 2002: 1224 pages

Mean and Dispersion of Ad Level: 2002-2012

.02

5.0

75

.12

5.1

75

.22

5

Ad

Pri

ce P

er

Re

ade

r ($

, de

flate

d)

2002 2004 2006 2008 2010 2012Year

Mean per-reader ad price 2002: 8.9 cents

Mean and Dispersion of Per-Reader Ad Price: 2002-2012

Data Preview: Market Revenue

20

120

220

320

Ad

Reve

nu

e P

er

Rea

de

r ($

, de

fla

ted)

2002 2004 2006 2008 2010 2012Year

Mean per-reader ad revenue 2002: $108

Mean and Dispersion of Per-Reader Ad Revenue: 2002-2012

75

80

85

90

95

Ad

re

ven

ue

/ (

ad

re

ven

ue

+ s

ubscrip

tio r

eve

nu

es)

(%)

2002 2004 2006 2008 2010 2012Year

Mean ad revenue share 2002: 80.5%

Mean and Dispersion of Ad Revenue Share: 2002-2012

How many subscribers go online-only?

Variable Mean Std. Dev. Min Max

Percentage of online-only subscribers (2012)

3.61 3.44 0.21 14.93

Estimation Results

Estimation: Reader Demand Theoretically, we are interested in

N = μ – ν (s + γa)

• For empirical specification,

njct = c0 – c1sjt – c2ajt + c3Xjt + Fj + Ft + Dct + εjct ;

• For interpretation,

N jt ct jct

c

M n

t

20 3 1

1

( ) - ( )

jt

ct jt j t ct ct jt jt

c c

cM c c X F F D M c s a

c

Estimation: Reader Demand

• Instrument: “Hausman-Nevo style” – (average) prices and

characteristics of magazines by the same publishers in other

markets (Hausman (1996); Nevo (2001)).

• Intuition: A publisher’s product prices in other markets are

correlated with this product’s price for cost-side reasons, but are

uncorrelated with this product’s quality (or taste shifters) b/c

they are in different market segments.

• For example, Condé Nast publishes both a general-interest

magazine (New Yorker) and a women fashion magazine

(Glamour).

• Kaiser and Wright (06, IJIO), and Kaiser and Song (09, IJIO)

use similar instruments.

Estimation: Reader Demand Coefficient Linear w/o

IV Linear w/o IV

Logit w/o IV Linear w/o IV

Linear w/ IV

Subscription price (in $10)

-.0026 (.0000

-.0005 (.0000)

-.0262 (.0095)

-.0005 (.0000)

-.0003 (.0000)

Ad volume (in 1000)

.0023 (.0000)

-.0005 (.0000)

-.1244 (.0128)

-.0006 (.0000)

-.0005 (.0000)

Gamma -0.8 cents 1.1 cents 5 cents Pre: 1.1 cents Post: .67cents

Pre: 1.5 cents Post:.85 cents

Year dummies No Yes Yes Yes Yes

Mag dummies No Yes Yes Yes Yes

Interactions? No No No post-2007*ad_level

.0002 (.0000)

post-2007*ad_level

.0002 (.0000)

Demographic controls

No No No No No

R-squared .02 .75 .49 .75 -

Off-range predictions

- 1 mag-year - 1 mag-year 6 mag-year pairs

Estimation: Advertiser Demand Theoretically, we are interested in

p = α – βa

• For empirical specification,

pjt = c4 – βajt + c5Yjt + Ft + εjt ;

• For interpretation,

4 5E(p )

t

jt jt t jtc c Y F a

Estimation: Advertiser Demand

Coefficient OLS OLS IV

Ad volume (in 1000)

.0074 (.0050)

.0129 (.0052)

-.0462 (.0219)

Year dummies No Yes (increasing and many significant)

Yes (increasing and many significant)

Avg subscriber demo controls

No No No

R-squared .002 .02 -

Model Prediction: Different Gamma

• In addition to the “minimal” specification where nu and gamma are the same for all magazines, we also ran regressions where gamma is allowed to differ across genres.

• This is to capture the idea that readers may value ads differently in different magazine categories.

• Kaiser-Song (IJIO 09) provides evidence on the heterogeneity of ad attitudes in German magazines.

• estimates in the next slides all come from IV estimation

• To compare with equilibrium relations

Model Prediction: Gamma vs. Ad Level

w. health

sci & tech

pop culture

parents

outdoor

men's

interior design

golf

w. general

lifestyle

general interest

food

w fashion

business

auto

.6.8

11

.21

.41

.6

Ad

leve

l (in

100

0 p

p)

-.1 -.05 0 .05 .1Gamma ($)

Fitted values

Ad level vs "gamma" by genre

Model Prediction: Gamma vs. Subscr Price

w. health

sci & tech

pop culture

parents

outdoormen's

interior design

golfw. general

lifestyle

general interest

food

w fashion

business

auto

11

.52

Su

bscri

ptio

n p

rice

per

issu

e (

$, d

eflate

d)

-.1 -.05 0 .05 .1Gamma ($)

Fitted values

Subscription price vs "gamma" by genre

Model Prediction: Gamma vs. Subscr Price

w. health

sci & tech

pop culture

outdoor

men's

interior design

golfw. general

lifestyle

general interest

food

w fashion

business

auto

1.2

1.4

1.6

1.8

2

Su

bscri

ptio

n p

rice

per

issu

e (

$, d

eflate

d)

-.1 -.05 0 .05 .1Gamma ($)

Fitted values

Subscription price vs "gamma" by genre

Model Prediction: Gamma vs. Full Price

w. health

sci & tech

pop culture

parents

outdoormen's

interior designgolf

w. general

lifestylegeneral interest

food

w fashion

business

auto

-150

-100

-50

05

01

00

An

nu

aliz

ed fu

ll p

rice (

$, d

efla

ted

)

-.1 -.05 0 .05 .1Gamma ($)

Fitted values

Full price vs "gamma" by genre

Fitness of the Parsimonious Model

What has changed in US magazine markets? Readers

• Reader demand, estimated time dummies are strictly decreasing.

• For example, estimated Year 2012 dummy is -.0019.

• This suggests that, for given demand shifters (e.g., mag characteristics) and on average, in 2012, each magazine loses 2 subscribers for every 1000 subscribes when compared to 2003.

What has changed in US magazine markets? Advertisers

• In advr demand, estimated time dummies are positive, increasing with many significant.

• For example, estimated Year 2010 dummy is .0408.

• (for given demand shifters and on average) in 2010, the advertiser w/ the highest wtp is willing to pay 4 more cents per eyeball to magazines, when compared to 2003

• Ad demand tilt:

Does beta vary before/after 2007?

Evidence that beta gets more negative after 2007.

This suggests that although top advertisers may be wtp more per eyeball, demand from bottom advertisers shrinks after 2007

Model Prediction: Subscription Price

** ( )

s2 2

a R a

From our theory, we have

Predicted Actual Ind. Mean

4.530859 1.42875 1.8989

2.974654 1.48049 1.8287

2.971645 1.30353 1.7283

2.099302 1.33409 1.6109

2.152304 1.154545 1.5025

1.779165 1.063827 1.4102

1.714479 0.919845 1.38

1.136203 0.974737 1.3782

1.836524 1.12143 1.347

1.271755 1.158 1.3448

0,5

1

1,5

2

2,5

3

3,5

4

4,5

5

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Details (w/ medium prices): Subscription price (per issue; $)

Predicted Actual Industry mean

Model Prediction: Average Ad Price From our theory, we have

𝑝 ∗ = (α + γ)/2

0,07

0,08

0,09

0,1

0,11

0,12

0,13

0,14

0,15

0,16

0,17

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Per-reader ad rate ($)

p_hat p_avg p_med

alpha gamma p_hat p_avg p_med year

0.145294 0.011 0.078147 0.1245 0.1169 2003

0.145294 0.011 0.078147 0.1291 0.1154 2004

0.176038 0.011 0.093519 0.1412 0.1317 2005

0.173867 0.011 0.092433 0.1387 0.1257 2006

0.173695 0.011 0.092347 0.1384 0.1291 2007

0.26412 0.0067 0.13541 0.1424 0.1344 2008

0.268009 0.0067 0.137354 0.1472 0.1349 2009

0.287053 0.0067 0.146877 0.1599 0.1358 2010

0.275114 0.0067 0.140907 0.1516 0.1374 2011

0.271215 0.0067 0.138958 0.1531 0.1379 2012

In sum: • Theory: U shape subscription price, increasing ad price

• Seems encouraging!

• Changes over a decade, changing business model:

• Subscription prices falling keeping readership constant

• Ad prices rising /constant and ad levels stepped down

(more ad finance in the mix?)

• Reader demand: ads as nuisance on average

• Ad demand: rising (per reader reached!)

• So saving sector by staunching reader loss

• Prices fitted OK: could back out costs

• Still to do: nuisance by mag genre; demographics in ad demand

Anecdotal Evidence

Time spent with different media

0 50 100 150 200 250 300

TV & video

Broadcast & satellite radio

Newspapers

Magazines

Internet

Average Minutes Per Day in U.S. (Source: eMarketer)

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

Magazine time!

36

37

38

39

40

41

42

43

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

(Source: eMarketer)

Share of magazine reading in total print-media time (unit: %)

1,5

2,0

2,5

3,0

3,5

4,0

4,5

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Share of magazine reading in total entertainment time (unit: %)

Where does ad money go?

Unit: % 2008 2009 2010 2011 2012

Magazine 18.1 17.1 16.8 16.1 16.2

Internet 6.7 7.9 7.6 9.1 8.7

TV 47.2 49.3 51.1 51.2 53.1

Radio 6.8 6 6.1 6 6.2

Newspaper 16.1 14.8 13.6 12.9 12

(Source: Kantar Media)

Media time vs. ad dollars in media

0

5

10

15

20

25

30

35

40

45

50

TV Radio Magazine Newspaper Internet

Consumer Time vs. Ad Money (2008)

% time spent

% ad money spent

0

10

20

30

40

50

60

TV Radio Magazine Newspaper Internet

Consumer Time vs. Ad Money (2012)

% time spent

% ad money spent

Different media effectiveness?

Points about others’ work

Rysman RES04 • US yellow pp: cross-sectional data with 476 directories.

• Directory is free, i.e., no subscription price. Homogenous product apart from # ads, so don’t need dummies for Directories. (cf. mags)

• Reader dd; Nested Logit; SHR, U = α2 lnA … so dd proportional to Aα2.

(so all have same ad-love, i.e., positive α2. However, since there is no subs price, they don’t provide the monetized value of ads. )

• Advr demand given by C-D at individual level (# pp. varies by advertiser), also a congestion effect (others’ competition) but can’t separately identify. Gets then downward sloping agg dd and inverts. Inverse demand Const elast -0.7 (includes downslope demand AND congestion!)

Notice MHA assumed, but then why, when there are multiple directories, could we still have such an effect from #readers? Reader elasticity on ad demand eq is 0.56. No discussion!

When we do that we get v similar 0.58 (given that ads are aggregated across the mags)

If there were constant returns in readers, this would be 1. Perhaps firms face diminishing returns from readers because say increasing marginal production costs. But then decisions across directories should be dependent upon each other. But, seems that the specification is used elsewhere with success: the price/ad/reader doesn’t work so well?

• Doesn’t engage equilibrium ad choice EXCEPT to infer (a la BLP: don’t estimate the eqm pricing relations because you don’t know costs!) marg cost of adding an advertiser to the directory, so we can’t really judge how reasonable it is …

• But, doesn’t effectively use eqm conditions to estimate equilibrium prices or ad levels, just the back-out

• [should we do likewise?? When we have the ads side down better? To do what tho?]

Fan AER13 US newspapers • US newspaper industry: panel data from 1997-2005. For advr dd, she

has 422 newspaper-year level observations. For reader dd, 8,947 observations of county-year level market.

• Ignores ad nuisance (from previous estimation, ad coefficient is small, negative and insignificant)

• Allows MHR (with dcm Logit rc including “both”),

but advertiser demand still Rysman monopoly type – except Rysman has an inverse demand (ad price on LHS), Fan has ad quantity on the LHS.

So not fully consistent (missed overlapping readers, etc)

• Newspaper firms sets both subscription and per-page ad prices

• Coefficients: reader-size elasticity of ad demand: 1.67 (ad demand goes up by 167% when the reader base doubles); price elasticity: -1.19.

• She uses eqm pricing to back out MC wrt A & N. reasonable?

• [Her main point/contribution is to endogenize quality charas of newspapers: non-ad space, # of staff for opinion section, # of reporters, the local news ratio, variety of sections ].

Wilbur MktgSci08

• US TV market: panel data on program viewing, prices and ads broadcasting for 50 US markets in 2003

• Dcm viewers, Logit rc

• Networks decide how many ads to air, but no subscription price for viewing

• Ads: linear inverse demands with eyeball demographics as controls

• Ad price has CRS to audience size: price per 30-sc goes up by $19 for every additional 1 million viewers.

• Gamma interacted w/ demogs; since there is no subs price, can’t provide a monetized ad nuisance cost.

• Eqm conds used to back off marg costs and for counter-factural analysis (ad-avoidance tech)

Concludes: “Not all eyeballs are created equal”

Lapo JAE07 • Italian newspaper industry: panel of 4 national

newspapers from 1976 -2003

• No ad nuisance

• Both sides dcm SHA SHR

• Reader nested logit; advr logit

• Newspapers set subs price and per-page ad price; both N and A have constant marg c.

• Advr demand: price elasticity -.91 to -.33 (different for diffrt newspapers)

• Also estimates a “collusive” model, where firms sets prices jointly; this model fits the data better!

Kaiser-Wright IJIO06

• Prescient! Themes in 2SM: who subsidizes who?

• 2-sided pos extys model, literal night-club application!

• 2sides Hotelling! (covered markets, only price diffs matter)

• Advertisers SH ! Like u= ρNR-a + konst (their notation: a is ad price)

• ad demand is linear function of sum of adprice diffs & reader diffs

• We have a different linear function, ours is (in their notation) n depends on a/NR

• In them, if price/ad/viewer const, advertiser demand goes up linearly (suggestion of ITV paradox type result) but in us it’s constant

• Split mags into 9 duopolies

Kaiser-Wright cont

• Why they don’t solve model?

• Don’t have to, they just want demands and then can back out mark-ups from focs (without jointly solving);

• We did similar with regress s on ad revenue and nuisance …

• (we could use this to back out costs? And ask the questions. We can also back out the costs from demand estimates and eqm equations

- How much are mags below MC (KW: answer? Tentative …)

- How much would they cost if no ads?

- If gamma is zero how many ads would there be?

- KW find that av across all mags, positive E/ad=… > E/content page =… ~!

KWright • Insts: prices of stablemate mags – like us

• Assumed covered markets on both sides

• Estimate separately ad demand and viewer demand

• Difference observations to remove year fixed effects

• 91 obs only, little price variation … how do they manage?

Kaiser Song IJIO09 • German magazine industry: panel of magazines in 6

categories; 7704 magazine quarter observations. 204 magazines covered.

• One side – logit reader demand, estimated separately for each magazine category (“genre”)

• On average, readers like magazine ads; moreover, in some categories, readers prefer ad pages to non-ad content pages!

• The degree of ad-loving varies across categories: largest in Women’s and TV.

• Their story: ads in Women’s and TV are more informative. They reach to this conclusion by randomly sampling magazines and count the number of “information cues” in each.

𝑝 ∗ = (α + γ)/2

Additional Notes

The p/v/ad puzzle

• Demand price/viewer/ad seems to decrease with the # of viewers. Why?

• Rysman had elasticity of 0.56

• Matthew ran (without mag dummies nor demographics) p = P/N on N and coefficient on N was negative (sig.), so we have a conc fn of P on N (not linear)

• So opposite of ITV premium

• What do others get?

U –shape ?

• Insert here if we have it!

• Tho we’d need to describe estimation of gamma first !

• Can we scatter-plot mags?

• Extra slides after the Concs (genres, on-lines…)

Recent patterns and interpretation

• (v.) loosely: falling subs price to 2008 then flat

Maintaining constant readership

• Rising then flat ad prices

Causing decreasing ad levels

Why the asymmetric response:

constant participation one side

constant price on the other?

Data pics: from MS;

• Subs levels flat, prices dropping

• Ad levels dropping, price flattened

• What combo of parameter changes can explain it?

• Bus finance model moving to more ad finance

• Competition online for both viewers and advertisers

(as the viewers go online)

[Maybe we need to model that too!]

Reader demand: first pass

• Regressions and results on gamma : breaking it down?

• Cf KW?

• Notice once we have gamma, we can look at the pricing eq:

s = k – R – a.gamma -- aside on how well this does! Or doesn’t

• So far on ad demand we have … can’t yet use fixed effects …