The effect of the Medicare Levy Surcharge on the Demand for

The effect of the Medicare Levy Surcharge on theDemand for Private Health Insurance in Australia

Olena Stavrunova (UTS)Oleg Yerokhin (University of Wollongong)

February 2012

Olena Stavrunova (UTS) Oleg Yerokhin (University of Wollongong)

The effect of the Medicare Levy Surcharge on the Demand for Private Health Insurance in Australia 1 / 37

Background: Health care system in Australia

• Mixed public-private health care system

• All residents are entitled to free public hospital treatment

• Private patients can bypass waiting lists and choose a doctor

• Private patients have to pay fee for doctor services plus costsof hospital accommodation, drugs, diagnostics, etc.

• Health insurance can be purchased to (partially) cover cost ofprivate treatment

• 43% are insured, 40% of elective surgery hospital admissionsare private



Policies to Increase Private Health Insurance Coverage• Since the end of 1990s several policies to increase PHI

coverage were implemented:I 1997- Medicare Levy Surcharge (MLS) - supplemental 1%

income tax on high income uninsuredI 1999- 30% insurance premium rebate ($2.5 billion per year)I 2000- Life Time Health Cover (LHC)- a system of entry-age

ratings: premium surcharge of 2 percent for every year that theinitial purchase is delayed after age 30.

• After the introduction of Medicare in 1984, PHI coverage hasbeen in decline. These policies increased PHI coverage from31% to 45% (45% increase)

• In 2008 threshold increased

• Currently proposed changes (FPHII Bill): means-testing PHIsubsidy and increasing MLS rate (three tier structure: 1%,1.25% and 1.5%)



Trend in Private Health Insurance Coverage

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

Hospital Treatment Coverage (insured persons as a percentage of the population)

Introduction of Life Time Health Cover from 1 July 2000 Commonwealth medical benefits at 30% flat rate

restricted to those with at least basic medical cover from September 1981

Introduction of Medicare from 1 February 1984

Medibank began on 1 July 1975. A program of universal, non contributory, health insurance; it replaced a system of government subsidised voluntary health insurance.

Introduction of 30% Rebate from 1 January 1999

Higher rebates for older persons from 1 April 2005

1 July 1997. A Medicare Levy Surcharge (MLS) of 1% of taxable income is introduced for higher income earners who do not take out private health insurance.

31 October 2008. Increase in MLS income thresholds, subject to annual adjustment.

Produced by the Private Health Insurance Administration Council, 2011



Literature

Because the timing of these policies overlap, is difficult to isolatetheir contributions to the total increase in PHI coverage in 2000Consensus in the literature: LHC had the strongest effect

• Butler (2002) and Frech et al. (2003): time series analysis onaggregate PHI coverage, attribute 11% increase to 30%premium rebate, the rest (34%) to LHC

• Palangkaraya and Yong (2007): individual-level data (NHS),attribute to LHC only 19%-33% increase

• No study attempted to isolate the effect of MLS (Small effectinitially?)

I The MLS could become more effective over time as fraction ofpeople with incomes exceeding the MLS threshold increased



Literature

There is international literature which seeks to establish how thedemand for PHI is affected by gov. incentives

• Subsidies (tax deductible PHI costs)I Gruber and Poterba (1994, QJE) - tax subsidies on the

demand for PHI by self-employed in USI Finkelstein (2002,JPubE) - reduction in tax subsidy for PHI in

QuebecI Emerson et al. (2001)- removal of tax subsidy for PHI for

elderly in UKI Rodriguez and Stoyanova (2004, HE) - tax reform affecting

the incentives to purchase PHI in Spain

• Mandates

Buchmueller et al. (2011, AEJ:Policy) - Employer Mandates inHawaii



Medicare Levy Surcharge

• Imposed on high income earners without PHI hospital cover

• MLS income thresholds from 1997-1998 to 2007-2008I Singles: $50,000I Families: $100,000 + $1500 per dependent child after the first

• MLS income thresholds since 2008-2009, with annualindexation

I Singles: $70,000I Families: $140,000+ $1500 per dependent child

• MLS rate: 1% of total taxable income

• High effective marginal tax rate at the threshold



How the MLS works: Single individuals, no children, 2008

47 48 49 50 51 52 5333

34

35

36

37

Before Tax Income, thousand dollars

Afte

r Tax

Inco

me,

thou

sand

dol

lars

0.70*(Before Tax Income)

With PHI (0.70*Before Tax Income)

No PHI (0.69*Before Tax Income)

50000 dollars

$500



Our study

• New dataset - income tax returns of the entire Australianpopulation (2008, 2010)

I Counts of individuals with and without PHI within $250 binsI Demographics: age (3 year intervals), genderI Family status (single no kids/otherwise)I Can study the effect on single individuals with no children

• A large number of observations around the MLS threshold -can estimate the MLS effect precisely using the regressiondiscontinuity type approach



Taxable income and PHI coverage (single, no kids), 2008

0 10 20 30 40 50 60 70 80 90 1000

0.5

1

1.5

2x 104

Taxable income, thousand dollars

Coun

ts o

f ind

ividu

als

Distribution of taxable income

0 10 20 30 40 50 60 70 80 90 1000

0.2

0.4

0.6

0.8

1


Prop

rotio

n wi

th P

HI

Relationship between income and PHI




0 10 20 30 40 50 60 70 80 90 1000

0.5

1

1.5

2x 104


Coun

ts o

f ind

ividu

als


0 10 20 30 40 50 60 70 80 90 1000

0.2

0.4

0.6

0.8

1


Prop

rotio

n wi

th P

HI




MLS incentives: Insurance choice

• m(c) utility from consumption

• I ∈ 0, 1 is private health insurance status (PHI)

• P is private insurance premium

• ξ is psychic cost of searching for PHI

• π is probability of illness

• τj is psychic cost of hospital treatment, τpb > 0, τpr = 0

• Cj is out-of-pocket cost of hospital treatment,Cpb = 0,Cpr = L · (1− I ) + 0 · (I )• L is such that a person without PHI will always choose public

hospital

• A person with PHI will always choose private hospital



MLS incentives: Reported Taxable Income

• Y is income

• E is the amount of tax avoidance

• Yr is reported income: Yr=Y − E

• T is MLS threshold ($50,000 in 2008)

• t is MLS tax (1%)

• a is per dollar cost of tax avoidance, a > t



Optimization

• Choose c, I and E to maximize expected utility u(c , I ):

u(c , I ) = m(c)− π · (1− I ) · τpb − I · ξ

subject to the budget constraint:

c ≤ Y − I · P − (1− I ) · t · (Y − E ) · ι(Y − E ≥ T )− aE

• Solution for E , conditional on I = 0:I E = Y − T + ε for Y ∈ [T ,Y )I E = 0 for Y ∈ [0,T − ε]

⋃[Y ,∞] where Y = a(T−ε)

a−t



Table: Utilities u|I and reported income Yr

Y < T T < Y < Y Y ≥ Y

u|I = 0 m(Y ) m(Y − a(Y − T + ε)) m(Y − t · Y )u|I = 1 m(Y − P) + η m(Y − P) + η m(Y − P) + η

Yr |I = 0 Y T − ε YYr |I = 1 Y Y Y

where η ≡ πτpb − ξ. If η ∼ F (η) ⇒ Pr(I = 1|Y ,P,T , t) andPr(I = 1|Yr ,P,T , t)



Income, reported income and Pr(I ): Y ∼ U , η ∼ N ,Pr(I |t = 0) = 0.19

4.95 5 5.05 5.1

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Y; Tens thousand dollars

p(I|Y

)

Probability of PHI, cond. on income

YbarTp(I|Y)p(I|Y,E=0)

4.95 5 5.05 5.1

x 104

-0.5

0

0.5

1

1.5

2

2.5

3

3.5

4

x 10-3

Y, Yr; Tens thousand dollars

p(Y)

, p(Y

r)

Distribution of income (Y) and taxable income (Yr)

p(Y)p(Yr)

YbarT

4.95 5 5.05 5.1

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Yr; Tens thousand dollars

p(I|Y

r)

Probability of PHI, cond. on taxable income (Yr)

p(I|Yr)

YbarT




35 40 45 50 55 60 650

5000

10000

15000


Coun

ts o

f ind

ividu

als


35 40 45 50 55 60 650.2

0.3

0.4

0.5

0.6

0.7


Prop

rotio

n wi

th P

HI





60 65 70 73 75 80 85 900

1000

2000

3000

4000

5000


Coun

ts o

f ind

ividu

als


60 65 70 73 75 80 85 900.2

0.3

0.4

0.5

0.6

0.7

0.8


Prop

rotio

n wi

th P

HI




Estimation

4.94 4.96 4.98 5 5.02 5.04 5.06 5.08 5.1

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Y; Tens thousand dollars

p(I

|Y)

Probability of PHI, cond. on income

Ybar

T

p(I|Y)

p(I|Y,E=0)

p(I|Y,t=0)

P|(Y=T)

• Effect of interest: for all Y>T

Pr(I=1|Y ,t=0.01)−Pr(I=1|Y ,t=0)

• Pr(I=1|Y :Y≥T ,t=0) not observable

• Estimate ∆Pr |(Y=T ):

∆Pr |(Y=T )≡Pr(I=1|Y=T ,t=0.01,E=0)

−Pr(I=1|Y=T ,t=0,E=0)

• Discontinuity in PHI due to MLSatY = T in the situation when taxavoidance is not possible



Estimation

1. Estimate Y and Y ≡ T − 2ε using taxable income distribution

2. Estimate Pr(I = 1|Y = T , t = 0,E = 0) by extrapolatingPr(I = 1|Y : Y < Y ) to Y = T

3. Estimate Pr(I = 1|Y = T , t = 0.01,E = 0) by extrapolatingPr(I = 1|Y : Y > Y ) to Y = T

4. Extrapolate ∆Pr |(Y = T ) to Y > T using variousassumptions



Possible discontinuity in sample selection probability



Solution

1. Using the whole sample (single and married males) estimate∆PrAll |(Y = T )

2. Assume that ∆PrM |(Y = T ) = 0

3. Therefore ∆PrAll |(Y = T ) = Pr(S |Y = T ) ·∆PrS |(Y = T )

4. Estimate Pr(S |(Y = T ))

5. Compute ∆PrS |(Y = T ) = ∆PrAll |(Y=T )Pr(S |(Y=T ))



Estimation of Y and Y

Using data on all males estimate

Ci=∑3

j=0 β0j Y

ji +

∑5k=1 γk ·ι(T−250·k<Yi<T−250·(k−1))

+∑15

k=1 αk ·ι(T+250·(k−1)<Yi<T+250·k)+εi

where Ci is counts of individuals in income bin i , Yi is themid-point of the taxable income bins, s.t.

∑5k=1 γk+

∑15k=1 αk=0

Y ≡ T − (k∗ + 1) · 250 where k∗ ≡ max{k : γk > 0}Y ≡ T + (k∗ + 1) · 250 where k∗ ≡ max{k : αk < 0}



Estimation of Y and YVariable Coefficient P-value

Y 3163.9 0

Y 2 -640805.5 0

Y 3 382942.4 0

γ1 7221.5 0γ2 1799.3 0γ3 762.1 0γ4 351.7 0.07γ5 66.3 0.731

α1 -2295.3 0α2 -1602.6 0α3 -978.0 0α4 -779.6 0α5 -745.6 0α6 -751.8 0α7 -499.5 0.01α8 -483.5 0.013α9 -502.0 0.01α10 -323.9 0.094α11 -207.4 0.282α12 -376.4 0.052α13 -250.0 0.195α14 -169.2 0.38α15 -236.1 0.221

const -27430.4 0

Y 48875Y 52625



Income distribution, males (single and married)10

000

1500

020

000

2500

0

30000 40000 50000 60000 70000income

Actual fitted distribution Counterfactual distributionCounts

Income distribution,2008



Estimation of ∆PrAll |(Y = T )

I Using the data on all males estimate

Ii = (β01 + β0

2Yi )ι(Yi <= Y ) + (β11 + β1

2Yi )ι(Yi >= Y ) + εi

where Ii is proportion of individuals with PHI in income bin i ,Yi is the mid-point of the taxable income bins, and

Y1 = 40, 875 Y = 48, 875 Y = 52, 625 YN = 64, 375

I Compute combined MLS effect:

∆PrAll |(Y = T ) = (β11 + β1

2 · 50125)− (β01 + β0

2 · 50125)

= 0.024∗∗∗



Estimation of ∆PrAll |(Y = T ).3

.4.5

.6

30000 40000 50000 60000 70000income

I- phirate2008I+



Robustness.3

.4.5

.6PH

I

40000 50000 60000 70000income

YhatT2008 phirate2008

.28

.3.3

2.3

4.3

6.3

8PH

I

38000 40000 42000 44000 46000 48000income

Fitted values endYhatT2008 phirate2008

.28

.3.3

2.3

4.3

6.3

8PH

I

38000 40000 42000 44000 46000 48000income


.28

.3.3

2.3

4.3

6.3

8PH

I

38000 40000 42000 44000 46000 48000income




Robustness.3

.4.5

.6PH

I

40000 50000 60000 70000income

YhatT2008 phirate2008

.4.4

5.5

.55

.6.6

5PH

I

55000 60000 65000 70000income


.4.4

5.5

.55

.6.6

5PH

I

55000 60000 65000 70000income


.4.4

5.5

.55

.6.6

5PH

I

55000 60000 65000 70000income




Robustness

1. The prediction error forPr(I = 1|Yne+5;Ynb : Yne ;Yne+5 ≤ Y ) is (−0.004,−0.002)Nobs = nb − ne + 1 = 33, Np = 8

2. The prediction error forPr(I = 1|Ynb−11;Ynb : Yne ;Ynb−11 ≥ Y ) is (−0.002, 0.003)(Nobs = nb − ne + 1 = 48, Np = 17)

3. The upper bound on the true effect is 0.024 + 0.002

4. ∆PrS |(Y = T ) = ∆PrAll |(Y=T )Pr(S |(Y=T ))



Estimation of Pr(S) and Pr(I = 1|Y = T , t = 0.01, S).3

.35

.4.4

5.5

.55

PHI

30000 40000 50000 60000 70000income

yh1 singlerate2008yh2

Single Rate

.2.3

.4.5

.6PH

I

30000 40000 50000 60000 70000income

yh1 phirateS2008yh2

PHI Rate, Single



Estimation of ∆PrS |(Y = T ) and ∆PrS |(Y > T )

• ∆PrS |(Y = T ) = ∆PrAll |(Y=T )P(S|(Y=T )) = 0.026

0.40 = 0.065

• Pr(I = 1|Y = T , t = 0.01,S) = 0.40⇒Pr(I = 1|Y = T , t = 0,S) = 0.335

• MLS increases PHI rate at Y = T by 20%.

• Extrapolate for i : Yi > T assuming constant effect per dollarof MLS tax:

PHICFi = Ci · (Prop(I = 1|Yi )− [0.01 · Yi ] ·0.065

500)

• Equivalent to assuming linear m(c) and η|Y ∼ U(aY , bY ) s.t.aY − bY = c



Extrapolating to Y > T.2

.4.6

.8PH

I

40000 60000 80000 100000income

phirateS2008CF phirateS2008

Actual and counterfactual PHI rate



Extrapolating to Y > T

Total counts, single males 3,076,275Total counts with PHI, actual 1,018,175Total counts with PHI, counterfactual 933,758∆ 84,417∆, percent 9%Overall PHI rate, actual 33%Overall PHI rate, counterfactual 30%Budget revenue, mln. 222.6



PHI rate (total), 2010.3

.4.5

.6.7

.8PH

I

40000 60000 80000 100000income

PHI rate,2010



Conclusion

• The effect of MLS in 2008 is relatively large at Y = T (butnot as large as expected)

• The total effect of MLS in 2008 is small

• The total MLS effect was probably even smaller when thepolicy was implemented in 1997

• The total MLS effect is probably even smaller when thethreshold increased to 73000 in 2008

• Future work: heterogeneity of the MLS effect by age andgender



Counts of individuals with PHI (single, male), 2008

35 40 45 50 55 60 65 705500

6000

6500

7000

7500


Coun

ts o

f ind

ividu

als

with

PHI

PHI coverage



Documents

The effect of the Medicare Levy Surcharge on the Demand for