CENTER FOR INSTITUTIONAL REFORM AND THE INFORMAL SECTOR

University of Maryland at College Park

Center Office: IRIS Center, 2105 Morrill Hall, CoIlege Park, MD 20742

Telephone (301) 405-3110

Fax (301) 405-3020

TAX AMNESTIES IN INDIA: AN

EMPIRICAL EVALUATION
Sizptember, 1995
Dilip Mookherjee and Arindam DasGupta
IRIS-India Working Paper No. 4

This publication was made possible through support provided by the U.S. Agency for International
Development in India, under Contract No. ANE-0015-B-13-1019-00 to the Center on Institutional
Rpfnrm and the Informal Sector (IRIS).
The views and analyses in the paper do not necessarily reflect the official position of the IRIS Center
or the U.S.A.I.D.
Authors: Dilip Mookherjee, Indian Statistical Institute, New Delhi, India and Arindam Das-Gupta,
National Tnsitute of Public Finance, New Delhi, India.

Tax Amnesties in India: An Empirical Evaluation

Arindam Das-Gupra
Dillp Mookherjee

Abstract
The purpose of this paper is to provide empirical estimates of the revenue impact df
Indian income tax amnesties between 1965 and 1993. A theoretical framework in a
companion paper examines the role of amnesties in allowing taxpayers to launder assets
accumulated by past tax evasion. Based on this theory, a dummy variable technique to study
the impact of amnesties on revenue is developed and applied to Indian data. Only the 1975
amnesty appears to have had a positive impact on revenue while other amnesties having
negligible or even negative effects. These results support the hypothesis that adverse
compliance effects of amnesties or falling pe.nalty collection can overwhelm direct gains from
an amnesty.

TAX AMNESTIES IX INDIA: AX EMPIRICAL

EVALUATION1
Arindam

Das-Gupta

National Institute of Public Finance and Policy, New Delhi

Dilip

Mookherjee

Boston University and Indian Statistical Institute, New Delhi

March. 1995

Introduction

Failure to impose sufficiently high penalties for tax offences is often cited as a cause of
widespread tax evasion, especially in developing countries. This view obtains support from
the theoretical analyses of Becker (1968) and Allingham and Sandmo (1972) which predict
that tax evasion is unambiguousIy

reduced by higher penalties, thereby suggesting that

enforcement policies should optimally involve the use of maximal penalties. In practice,
penalties for most offenses tend to be non-maximal. Indeed, a common practice in many
countries is to offer amnesties for past offenses in an effort to encourage voluntary disclosure
of past tax evasion. Amnesties have been used by both developed and developing countries
in recent years. For instance, during the 198Os,
in more

tax amnesty programs have been employed

thtin half the states in the United States. besides Belgium. France, Ireland and

Italy. Amongst developing countries? they have been employed in Argentina, Bolivia, Chile,
Colombia, Ecuador, Panama, Peru, Mexico and the Philippines more than once in the recent
past.
India has also offered a large number of amnesty programs in past decades: in 1951,
1965. 1975,

1980. 1985, 1986 and. most recently, in 1991. Since 1975, therefore, they have

been offered frequently and predictably: once every five or six years. Section 2 of the paper
discusses the nature of these programs in more detail. These amnesties have formed part of
a general climate of weakening enforcement of the Indian income tax. Other manifestations
include a Iower intensity of prosecution effort for tax evasion. and the introduction of various
forms of out-of-court settlement of evaded income taxes due from large taxpayers.3
The purpose of this paper is to provide empirical estimates of the revenue effects of
various (temporary) amnesty schemes offered in India since 1965. The approach used folIUV-s

the theoretical framework laid out in a companion paper (Das-Gupta and hiookherjee

(1995)),

which examines the role of amnesty programs in allowing taxpayers to launder

previo~zsl~ undisclosed (black) assets. Section 3 of the paper provides a broad summary
the main

of

results of this approach. These results stand in contrast to a number of prevailing

views concerning the usefulness of temporary amnesty programs. The differences stem from
the indirect revenue effects that arise from the effects of the amnesty on filing behaviour
1

and on subsequent collections from penalties. Being spread out over a number of years
both preceding and following the amnesty year. these indirect effects tend to be less visible
than the direct collections from the amnesty. These indirect effects are. however. typically
adverse. and can outweigh the direct effects.
Empirical results pertaining to the revenue effects of the amnesties are presented in
Section 4.

Utilizing aggregate data for the period 1965-66 to 1992-93:

these effects are

estimated after controlling for other important influences on aggregate income tax revenues.
The temporal specification of dummy variables used to capture the effect of amnesties draws
on the theoretical predictions described in Section 3.
Based on these estimates. a number of broad conclusions concerning the effect of amnesties
in India emerge. The three amnesties that took place before the mid-1980s were arguably
unanticipated by taxpayers. Of these, only the 1975 amnesty lead to significant revenue
gains of the order of 25% of annual revenues. The 1965 and 1980 amnesties had negative
overall effects on revenues, even during the amnesty year itself.4
Following 1980, amnesties have been periodic and therefore anticipated. Their indirect revenue effects were adverse. and outweighed direct amnesty receipts. Nevertheless:
the overall revenue effects of amnesty programs from 1980 onwards have been small relative to aggregate income tax revenues. with estimated effects frequently turning out to be
statistically

insignificant.

These results provide support for the hypothesis that negative indirect compliance effects
or falling penalty collections can overwhelm the direct receipts from amnesties. This is
particularly truewhen amnesties are offered at predictable intervals. In light of this, and
the fact that the revenue impact of amnesties appear to be shrinking progressively, Indian
tax administrators would be well advised to resort to alternative methods for improving
enforcement of the income tax.

2 Tax Amnesties in India

This section describes the main features of various amnesty schemes offered in India since
1950. excluding permanent amnesties.5

This serves to provide background informatlon

for

the analysis of later sections.

Voluntary disclosure schemes were offered during the years 1951, 1965 (two schemes)?
1975: 1985 (extended to 1986) and 1991. These schemes provided taxpayers immunity from
penalt>- and prosecution in exchange for disclosures of income which had evaded tax in the
past. In 1980 the government introduced a Bearer Bond Scheme. Funds used for their
purchase were promised immunity from investigation or prosecution. The bonds had a
maturity

period of ten years, with a cumulative interest of 20 percent payable at the;end

of the period. Government sales of these bonds are not reflected in income tax revenue
receipts, unlike other amnesties. The invested amount was not subject to income. wealth or
gift taxes. Bagchi (1981) calculates that the interest rate was below market rates by almost
exactly enough to cancel out the tax benefits, so there was no implicit subsidy on these
bonds for the average taxpayer. Table 1 describes the main features of the various nonpermanent amnesty schemes between 1951 and 1986; complete detaiIs of the 1991 amnestS
are not as yet available.
The objecti\-e of the Government in introducing these amnesty schemes appears to have
been the curbing of tax evasion and to enable taxpayers to disclose their hitherto unreported
(black) assets. As the report of the ComptroiIer

and Auditor General on the 1985 amnesty

scheme stated:
fCo induce the tax-payers to respond and declare larger incomes? the Government further adopted a lenient and sympathetic approach as regards the penal
proceedings and offered amnesty to repentant tax-payers to reform themselves
and to declare their true income/wealth..... Salient features of the Scheme were
- Sole objective of curbing tax evasion and unearthing the considerable amount
of black money that is vitiating the nations economy - No probe to be made
into past assessments and immunity afforded from interest, penalty and pros3

ecution on true and full disclosures of income or wealth and prompt payment
of tax - No undue benefit to repentant tax-payer vis-a-vis honest tax-payer...
(CAG (1990. p.v)
The stated objectives do not include notions of signalling future enforcement resolve, or the
need to

provide

benefits of insurance or consumption smoothing to taxpayers (objectives

stressed by a number of amnesties in other countries, as well most existing theoretical

treatn~ents).6

The scheme amounted to an admission by the government of its failure to

curb tax evasion in the past. and constituted an attempt to provide taxpayers with the
opportunity to launder their black assets

into

white.

It is evident from Table 1 that amnesty receipts have formed a shrinking fraction of
revenue receipts, falling from about a fifth of total revenues in the 60s and

70s

to one-tenth

or less during the 1980s. In this sense, the direct returns from the amnesties have been quite
meagre in comparison with the

mm&e estimates of black money or tax evasion of Acharya

et. al. (1985) or Virmani (1986).

D ue

to the poor response, many government commit-

tees have

advised against their use. starting with the 1958-59 Direct Taxes Administrative

Enquiry

(Tyagi) Committee. the 1971 Direct Taxes Enquiry (Wanchoo) Committee, and

going on to Acharya et.nZ. (1985). _Uevertheless:

the government continues to use amnesty

schemes.. most recently in 1991 when four schemes were announced: three aimed at domestic evaders. and one designed to attract foreign capital inflows. Revenue estimates for the
1991 schemes are not currently available.
Of the various amnesties offered, the 1975 scheme coincided with the declaration of an
internal Emergency, and was associated with a curtailment of civil liberties, stepped up
search activity, and an increase in the rate of conviction for tax offenses. Given the ten year
gap between this and the previous amnesty: it appears reasonable to suppose that the 1975
amnesty was unanticipated. It may also have been interpreted as a prelude to stepped up
enforcement.

TABLE 1: TEMPORARY AMNESTIES IN INDIA

I
Duration
Year ! Authority For
Tax
Xurnber.
Income
Revelme
!
c The Scheme
Rate
of
DiscloYield
Cases

sed

( 0 0 0 )
1951

1i.a.

r1.a.

May-Oct.

Yield As
% of
Years

(Rm.) (Rscr.)

Revenues

20.9

70.2

10.89

2.0

52.18

114.2

145

19.45

245.6

746.07

249

20.5

400

160*

10.6*

2940.37

388.03

7.2

1951
1965

Finance Act

60%

March-Ma-y

1965
1965

1975

1980

1i.a.

Aug 1965-

1065

Voluntary

30.8

18.5t

1965

Finance Act
(no.2)

Mar 1966

1 25-

Disclosure
Act 1976
1
Special Bearer :

Ott 1975

60%
I
*

**

Nov 1985

Bond Scheme
1981
198ri

Finance

-1.986

1985

Act

1539.9

Mar 1987
Note:

*: See text for explanation; 1 :Aggregate of the two 1965 schemes;

**: SlifJject to ~lormal tax rates, but waivers of interest gramed. S o u r c e : C A G (1990).

The 1965 amnesty was unlikely to have been anticipated. with just one amnesty scheme
previously offered in 1951.

There is no reason to suppose that the 1980 amnesty was

anticipated either. given the unusual circumstances of the previous one. The end of the
Emergency in 1977 was followed by a significant diIution

of standards of enforcement. This

suggests that the introduction of the 1980 amnesty induced taxpayers to associate them with
weak enforcement resolve and therefore expect amnesties to be introduced periodically. A
policy of repeated amnesties could be favoured by a government. that gave importance to
short term revenue gains from amnesty disclosures over long term revenue losses. These
expectations came to be realized. of course, with amnesties offered in 1985-86 and again in
1991.s
The empirical modelling

therefore proceeds under the assumption that the 1965, .1975

and 1980 amnesties were unanticipated. Uoreover. it is supposed that the introduction of
the 1965 and 197.5 amnesties did not lead any fresh amnesties to be anticipated in the future.
whereas the 1980 amnesty did lead to such an expectation. Consequently, the 1985-86 and
1991 amnesties are assumed to have been anticipated.
One additional feature of these amnesties deserve to be noted: the tax rate applicable
to disclosures were typically below the maximum marginal tax rate on ordinary income
disclosures prevailing at the time. In 1965. the maximum marginal rate was 65(rc, with
an additional surcharge of around 2%. The amnesty tax rate of 60% was therefore lower
than this rate.

In 1975. the top marginal tax rate was 77% substantially higher than

the maximum rate of 60% applied to assets declared in the amnesty. For taxpayers whose
disclosures were large enough to fall in the highest income bracket, therefore, these amnesties
amounted to allowing a concessional tax rate: relative to taxes liable for earlier (or even
contemporaneous) disclosures. hforever.

the 1980 Bearer Bond Scheme effectively amounted

to a zero effective tax on the amounts disclosed. Finally. the 1985-86 amnesty levied tax
at the usual rates, but waived interest on the delayed payment. This was tantamount to
allowing a concessional tax rate. lo Consequences of these concessions are discussed below.

3 Theoretical Framework
This section outlines a framework for analysing the effects of a temporary amnesty in inducing voluntary disclosure of black assets to increase government revenues. the stated
objective of the Indian amnesty schemes. The main conceptual elements necessary to evaluate the disclosure and revenue effects

of

amnesties are introduced. This is followed by a

summary of predictions concerning these effects that will prove useful in interpreting the
elnpirical results. The theory is developed
Mookherjee

furmally in a companion paper (Das-Gupta and

(1995)).

Black asset.s represent accumulation

of past mcomes which were not voluntarily disclosed

by a taxpayer. nor discovered by tax authorities. These assets are typically invested in the
underground

economy

where

sources

of in\-estible funds are not usually investigated (such

as construction, real estate purchases or movie-making). Consequently, the growth of these

sectors

is r&ted to the accumulation of black ass&s

among

citjlenn

Tn contrast,

white

assets are invested in legal sectors such as deposits with financial institutions, or shares of
Iegal companies held. The fact that the two kinds of assets are invested in different sectors
has two important implications. First, incomes from white assets are easier to monitor
b!- tax authoribies.

Second. the rates of return on black and white assets typicahy

differ.

olving to Iimited mobility of assets between the two sectors. While empirical evidence on
relative rates of return are not available. casual evidence suggests that rates of return are
substantially higher in the black sector. though accompanied by higher levels of risk.
Incentives for voluntary disclosure of incomes depend on the probability of undisclosed
income being discovered by tax authorities. and the level of penalties imposed for tax
evasion.

Unlike white incomes. citizens usually have no incentive to voluntariIy

disclose

bIack assets or incomes from these assets in the normal course. The reason is that the
authorities would impose penalties on these disclosures, as they are tantamount to admission
of past tax evasion. Penalties differ little from those that would be imposed if the assets
Ivere not to be disclosed. and were discovered by the tax authorities instead. The citizen
therefore has no incentive to disclose black assets as this would simply increase rhe likelihood
of incurring penalties for past tax evasi0n.l
7

.i\n amnesty lowers the penalty rate on voluntary disclosures of black assets, relative to
tire penalties imposed for discovery of these assets by.tax authorities. This promotes the
cirizen!s

incentives to voluntarily disclose these assets. The disclosure decision trades off

the certainty of having the pay the amnesty fine rate (consequent on participating in the
amnesty). v,-ith the probability of having these assets discovered later by the tax authorities
and having to incur higher penalties (consequent on not participating).
If the taxpayer participates in the amnesty, black assets disclosed will thereafter be
converted into white assets. In other words. the amnesty serves as a device for the taxpayer
to l~un.&r black assets into white. Owing to the intrinsic differences between white and
black assets described above, this will have an impact on income flows and tax c.ompliance
in the future. To the extent that rates of return on white assets are lower than they are on
black assets. income flows accruing to the citizen in the future will be diminished. Forever,
since income from white assets are more easily detected by tax authorities, tax compliance
in the future is Iikely to be enhanced.
The idea of encouraging voluntary disclosures of black assets via a temporary amnesty
may appear sensible. on the grounds that it would increase revenues directly (in the form
of the taxes and penalties owed on the declared assets) during the amnesty as well as in
later years by switching income
complian1.e.

flows

of citizens from black to white thus promoting greater

Stella (1990), for instance, suggests that an amnesty may even be a necessary

precondition for t.he success of a program of stepped up enforcement aimed particularly at

discox-cry of undisclosed white incomes.r2
Countering these beneficial effects is the possible adverse impact on tax compliance in
pre-amnesty years, if the amnesty happens to have been anticipated by citizens. In such
instances, they will be induced to engage in a form of intertemporal substitution with respect
to their voluntary disclosures: disclose less in pre-amnesty years. so as to accumulate more
black assets. which are subsequently disclosed in the amnesty. This will generate an adverse
revenue impact in years immediately preceding an amnesty which is correctly anticipated.
Part of the revenue gains accruing in the form of receipts from the amnesty will thus tend
to be illusory. and will be mirrored by revenue losses in pre-amnesty years. This forms the

basis of the view that amnesties are more effective when they are not anticipated in advance
by citizens. and when they do not lead further amnesties to be expected in the future. Stella
(1990) argues that for this reason amnesties ought to be used sparingly - perhaps on a
once-and-for-all basis - otherwise the government will find it hard to credibly commit to
not offering additional amnesties in the future.
Nevertheless! a detailed examination of the preceding arguments suggests that they are
simplistic and ignore a number of additional compliance and revenue effects of temporary
amnesties. These supplementary effects are typically indirect and less visible compared to
the direct effects outlined above. Forever, they usually run in an opposite direction to the
direct effects and under fairly natural conditrons

can overwhelm the latter.

The main issue is the incentive of a citizen with black assets to participate in an amnesty.
AS explained above: the main motive is to avoid having to incur higher penalties if these
assets are not disclosed in the amnesty and are discovered later by tax authorities. The
amnesty penalty rate f applicable to the tax due on these assets must therefore be low
enough to induce participation. relative to the probabihty

that they will be subsequently

discovered (in the absence of disclosure) and the penalty rate 7 applicable in the latter
instance. Suppose that monetary penalties are imposed rather than prosecution or imprisonment: this is a realistic assumption in the current Indian context.13
It follows from this consideration that
am.nesty:

the govern.m.ent

there

is an.

adverse ~$Jwt of pwticippntion.

in, the

will forego the revenues that it would have earned from. penalties

in the future had th.e citizen. n.ot participa,ted

in th.e am.nesty,

and

if black assets were subse-

quently discovered. In other words? the increased revenues from the amnesty disclosures will

be followed by a subsequent decline in additional collections from penalties. compounding

or settlement fees.
How do these two opposite effects on revenues compare with one another? Consider
first the case where the pre-tax rat.es of return on black and white assets are the same and
the citizen is risk-neutral. This assumption implies that participation in the amnesty does
not affect

the stream of

tots1

income accruing

to the citizen,

before taxes and penalties

are accounted for. Then the citizen participates in the amnesty only in t.he expectation of

lowering total

must

experted (tax and penalty) payments

to

the government. The consequence

be that total expected receipts of the govern.men.t

m.ust fall in

th.e long run.. This

appear counterintuitive in circumstances where discovery of black assets is especially

may

unlikely, as that would suggest that the expected reduction in subsequent penalty collections
(following amnesty participation) will be small. What such an argument overlooks is that in
such circumstances the amnesty fine rate f must be concomitantly lower in order to induce
participation in the amnesty.
Now suppose that the expected rate of return on black assets exceeds that on white
RSSC?~R.

Then amnesty participation must incur an additional loss to the taxpayer, arising

from a reduction in total pretax income which is the result of switching assets from black
to white. To induce participation. the amnesty rate must be even smaller: thereby reducing
the total revenue effect of the amnesty even further.
-A third additional consideration arises when the amnesty is a sale, in the sense that
the effective tax rate on these disclosures is less than the tax rate apphcable

to normal

white income disclosures (2. c., the amnesty fine rate f is negative). It has been noted in
the preceding section that this appears to have been the case for most Indian amnesties.
This creates an incentive for habitual fiIers to mis-file, in the following sense: instead of
declaring their white income jn the customary fashion in the amnesty year. they declare it
as .pre\-iously undisclosed income in the amnesty in order to avail of the concessional tax
rate. Since the source of amnesty disclosures are not investigated by the tax authorities:
such n&-filing behaviour

is difficult to detect. The consequence is a net reduction in total

voluntary collections during the amnesty year. Some amnesties which are sales prohibit such
djversjons

explicitly. but the Indian amnesties have been somewhat ambiguous in this regard

and do not appear to have successfully deterred such cases of mis-filing.14

Consequently,

some of the receipts of an amnesty may actually represent diversions from taxes that would
have been voluntarily paid in the absence of t.1~~ awlesty.
Why might a government ever want to allow an amnesty to be a sale: i.e., offer concessions on the tax liability, over and above immunity from penalties and prosecution? The
main motive may be to ensure that participation in the amnesty is not too limited. which

would make it appear to have failed. It can be shown that habitual non-filers of income
tax will participate in an amnesty only ij it is a sale. In other words: an amnesty which is
not a sale will only attract habitual filers of the income tax who by their very nature are
unlikely to have accumulated much black assets. In order to widen participation to include
non-filers, an amnesty must therefore be a sale. The cost of this is the direct reduction in
amnesty revenues from filers who would have partxcipated
addition

in the amnesty m any case: in

to the revenue losses arising from mis-filing by habitual filers.

The preceding arguments indicate why the direct receipts from an amnesty ma>onI-

provide

a very limited perspective on their aggregate revenue effects. Note that the arguments

apply to amnesties which are unanticipated.

Indeed, such an amnesty may :appear to

be a success. judging by the direct receipts, as well as by the boost to voluntarily filed
taxes in years following; ~lle anulesty (owing to the induced asset laundering).

The losses

resulting from reduced penalties will only appear in the future and will typically not, be
i&ntificd

OS a

cost

attributable to the amnesty Lnnsns

dll@ t,n mis-filing during

the amnesty

year also cannot be identified. unless all other factors that help determine the level of tax
revenues are explicitly controlled for. An econometric analysis that carefully controls for
other determining influences and examines the revenue effects in a number of years following
the amnesty. is required in order to estimate the indirect effects and contrast them with the
direct revenue receipts. This is the objective of the following section.
Some qualifications to the preceding results concerning the adverse revenue effects of
unanticipated

amnesties need to be noted.

First, if rates of return on black assets are

lower than on white assets, the asset laundering effect of an amnesty may increase revenues.
Second. if taxpayers are risk averse, then the amnesty has the feature of offering insurance to
taxpayers against the risk of their black assets being discovered later. The government could
conceivably increase the amnesty fine rate in order to capture the associated risk premium,
\\:ithout

jeopardizing

participation.

Third.

if

imprisonment

rather

than

monetary

penalties

result in the event of discovery of black assets by tax authorities, then an amnesty may
result in an increase in total revenues owing

to

an effective conversion of nonpecuniary into

pecuniary penalties. Amnesties may also lead to a saving in the cost of prosecuting offenders
tvho

would have been detected and prosecuted in the absence of amnesty participation.

HOW

js tthe analysis modified when an amnesty is anticipated? Somewhat contrary to

con\;entional wisdom. anticipation of an amnesty may have a less adverse impact on revenue
than an unanticipated amnesty. Citizens do indeed engage in intertemporal substitution
with respect to their payments to the government when they anticipate the amnesty: they
accumulate black assets faster in pre-amnesty years and then disclose them in the amnesty.
The effect of this is to switch their asset portfolios in favour of black assets in pre-amnesty
years. If black rates of return are higher than white rates the:effect of the amnesty is
to increase the before-tax total income of the taxpayer. Some of this increase can be
appropriated by the government, in the form of a higher tax rate imposed on assets disclosed
in the amnesty. As a consequence. it is conceivabIe
generates a favourable

that an amnesty which is anticipated

effect on aggregate revenues! whereas it would have an unfavourable

effect if it were unanticipated. Nevertheless, this can happen only if white incomes that
are not reported in pre-amnesty years in anticipation of an amnesty are sufficiently large
relative to total black assets. Furthermore. it can be shown that an anticipated amnesty
will have a negative overall revenue effect if the probability q of discovery of black assets by
tax authorities is smalls even if the proportion of white assets switched prior to an amnesty
is substantial relative to black assets.
Predicted Am,n,ssty

Effects This section concludes with a summary of predictions con-

cerning the timing of revenue effects of amnesties that are respectively, unanticipated and
anticipated. These will be useful in setting up the dynamic specification of the revenue
regressions in the following section and in interpreting the empirical results.
For an amnesty which is unanticipated:
1. Prepaid taxes at the amnesty date increase by the amount of amnesty receipts, or by
less if the amnesty induces some taxpayers to mis-file. With sufficient mis-filing.

they

may even fall.

2. Prepaid taxes following the amnesty also increase.
3. Addit.ional
dates.

collections from penalties decrease at both amnesty and post-amnesty

because of the laundering of black assets which would otherwise have been
12

-- - --

-G--69

----- _----_ -

@- a--@

--f&-;--

---- ---- ----- _-_-__ -_--

-.

49
a -_---__ ---_ -VW_ - -. _-----c3
a, @@
@

subject to penalties or settlement fees on discovery.

4. Consequently. total collections (the sum of prepaid taxes and additional collections
from penalties) increase by less than direct amnesty receipts at the amnesty date and
may even fall (owing to mis-filing). At later dates, also, total collections increase less
than do prepaid taxes, and could conceivably decrease.
5. If the pre-tax gross return (not adjusted for risk of disco.very)

on black assets is

not lower than on white assets; t,hen the overall revenue effect aggregated across the
amnest,y and post-amnesty dates, is negative.
If the amnesty is anticipated. then we would expect the same results to hold during and
after tbe amnesty. In addition:
1. Pm-paid
effect

on

collections in pre-amnesty years fall. Additional collections rise! but the

aggregate collections in pre-amnesty years cannot be predicted.

2. TI~c ox-era11 revenue effect: agsegated across a11 dates. is expected to be negative.r6

INSERT FIGURES 1 AND 2

These results are illustrated in Figures 1 and 2. for the case where there are long lags
invob-ed

in collecting penalties associated with tax offences

discovered in any year and losses

due to mis-filing are not large. They assume that all other factors are unchanging over the
period under consideration. so that revenues would be stationary in the absence of the
amnesty. For an unanticipated amnesty: therefore. revenues are not affected in pre-amnesty
years. JVith some (but not excessive) mis-filing, and delayed penaltres

or settlements of tax

evasion cases. revenues increase in the amnesty year, but by less than the direct amnesty
receipts. Revenues also increase Immediately following the amnesty, owing to ~hr. switch
from black to white assets by taxpayers. which promote compliance. They fall eventually.
however. as the reduced penalties and settlement fees reflect the lower stock of black assets.
For an anticipated amnesty. there is the additional effect of a drop in pre-amnesty revenues
olving to the intertemporal substitution in compliance.
13

These figures also depict a situation where the introduction of the amnesty does not signal information to taxpayers concerning future enforcement resolve of the government. Such
signalling effects could either serve to step compliance up or down following the amnesty,
depending on the precise context. In the Indian context. it may be expected that the 1975
amnesty served to signal tougher enforcement in the years ahead, while the others had the
opposite effect. Accordingly, one would expect the amnesty year and post-amnesty effects
of the 1975 amnesty to be higher than depicted in Figures 1 and 2, and lower for all the
others.

4 Empirical

Analysis

This section presents estimates of the revenue effects of various amnesties in India since
19~5, after controlling

for other determinants of income tax revenues l7 The estimates are

based on yearly data for income tax revenues for the period spanning financial years 1965-66
to 1992-93. Unfnrtnnately,

a breakdown of revenues between prepaid taxes and additional

collections from penalties is not available for most of this period. This prevents us from being
able t.o test some of the detailed predictions described at the end of the previous section,
in which effects on prepaid taxes differ qualitatively from those on additional collections.
No information is available concerning the length of Iags in collecting penalties associated
\vith tax offences. Consequently, the collection figures in any given year correspond to the
sum of prepaid taxes for that year and additional collections associated with tax evasion in
a number of previous years.
The revenne effects of various amnesties are estimated by the coefficients of various
variables corresponding to pre-amnesty, amnesty and post-amnesty years. As explained
in Section 2. the 1965-66,

1975-76 and 1980-81 amnesties are treated as unanticipated

amnesties. unlike the subsequent ones. Accordingly, for these three amnesties we do not
have any dummies corresponding to pre-amnesty years and instead have one dummy for the
amnesty year effect and another for post-amnesty effects. With the 1985-87 and 1991-92
amnesties being treated as anticipated. pre-amnesty effects also ha\-e to be incorporated for

14

these. Since following 1980-81 successive amnesties were spaced only five years apart. the

periods

1981-82 to 1984-85

and 1987-88 to 1990-91 can be viewed as combining the post-

amnesty effects of the preceding amnesty and the pre-amnesty effects of the succeeding
one. Consequently. in one specification used below, we simply include a composite dummy
variable for the interamnesty years 1981-82 to 84-85 and 1987-88 to 1990-91. In the other,
less parsimonioust specification we treat the two years following an amnesty as embodying
a post-amnesty effect and the subsequent two years a (possibly distinct) pre-amnesty effect
in anticipation of the next one.
In estimating the effects of the amnesties, the following infiuences on tax revenues are
controlled for: non-agricultural gross domestic product (NAGDP); the inflation rate (INFL);
the effective marginal income tax rate averaged across different income groups (AMTR); the
exemption Iimit for incomes (EXEMJ below which taxes are not liable; and two indicators of

enforcement effort: the value of assets seized during search and seizure operations (SEIZ),
and an index of assessment activity (ASS). * Other potentially important variables such
as prosecutions in tax evasion cases or the conviction success rates for the Income Tax
Department 1 turn out to be jointly

st&sCic& i~A&Gcant.lg

The dummy variable approach is subject to certain shortcomings. First. the method
itself imputes the ent.ire

effect

of all influences on tax revenues to amnesties. apart from

variables explicitly controlled for. To the extent that variations in enforcement effort are
not adequately captured by the variables included in the regression. they could simply
an omitted variable bias. For instance. the effects aatributed
s:inlpl>~ ieffect the cffcct

of

reflect

to the 1975-76 amnesty could

introduction of the internal ErnerEwwy, which signall~d

stronger

enforcement resolve. A second problem stems from the fact that assessments and revenue
collections for any one year arise partially out of taxes filed for earlier years. This can lead

to bias in the estimation of temporal effects of amnesty schemes.31

hloreverT
pation
used.

the theory says littIc about the number of years that the pre-amnest+

ant&i-

and post-amnesty laundering effects persist. Given the shortness of the time series
there is a need to URN as parsimonious a specification as is permitted by the data. in

order to inlprove the accuracy of the estimates and hypothesis tests. This creates the need

15

to experiment

u-ith

different specifications of the temporal structure for these effects. 111 this

respect a general-to-specific methodology was adopted:, starting with the most unrestricted
specification permitted by the data, successive restrictions were tested. in order to arriye
at the most parsimonious specification available: i.e., where further simplifying restrictions
are rejected.
The completely unrestricted specification (referred to as Specification U below) allowed
for distinct Z-period pre and post amnesty effects, and different amnesties to have disparate
effects. The most parsimonious specification (referred to as specification P below) uses a
single dummy for all interamnesty srears following 1980. thereblr assuming that the revenue
effect was equal across al1 these years. In addition, the amnesty year effects of the 1985-87
and the 1991-92

amnesties are also assumed to be equal. Coefficient estimates correspond-

ing to both specifications are reported in Table 2 beluw.

the results of level and first-difference (Engle-Granger)
suits do not vary rlmh between these apccificstions.

Fur each
regressions.23

specification we report

Fortunately, the re-

suggesting that inferences

concerning

significant influences do not spuriously arise from a common time trend.

4.1 Empirical Results

Table 2 presents estimated coefficients for both parsimonious (P) and unrestricted (U)
specifications.
interpretation.

-411 variables are measured in logs,

SO

the coefficients have an elasticit!

In particular. the coefficient of any dummy variable can be interpreted

RS the percentage change in revenues that can be attributed to the corresponding period.
The special circumstances of the 1980 amnesty neeiI to be kept in mind in iinterpreting
these coefficients: the direct revenue receipts from this amnesty are not included in the
revenue series used. unlike other amnesties. Hence the coefficient for 1980 represents only
the indirect effects of that amnesty during the amnesty year. Government receipts from

the

Bearer Bond Scheme must be added to the indirect revenue effect implied by the coefficient
to obtain the total effect of the 1980 amnesty. Table 3 presents absolute revenue effects
associated with different amnesties with the 1980 estimate adjusted to reflect direct effects
of the Bearer Bond Scheme.
16

TABLE

I
I

2; COEFFICIENT ESTIMATES

Specification P

Specification U
First

Level First diff.

LPWl

- 0 . 3 6 -0.38*

-0.61 -0.53

1966-7

- 0 . 2 7 -0.29*

-0.45 -0.40*

1975

0.30*

0.27

0.25

0.28

1976-7

0.04

0.05

-0.01

0.03?

1980

-0.18 -0.15*

-0.16*

-0.16*

1981-2

-0.15

-O.ll?

-0.04

-o.oEi*

1983-M

-0.15*

-O.ll?

-0.00

-0.07+

1985-6..

-0. ll?

-0.04

0.06

0.02

1987-88

-0.15*

-O.ll?

-0.00

-0.07+

1989-90

-0.15*

-O.ll?

-0.07

-0.10*

1991

-O.ll?

-0.04

0.11

u.01

1992

-0.15*

-0. ll?

-0.07

-0.14*

YAGDP

0.?34*

0.89+

0.55

0.70*

INFL

-0.91 -1.00*

-1.48

-1.11*

ASS

0.59

0.6?*

1.00+

0.88

SE12

0.03

0.03*

0.06

0.05*

EXEU

-o-59*

-0 a*

-1).69t

-0.55*

AhITR

-0.65

-0.49

-0.80+

-0.70*

1965

Error

Diff.

-0.96*

Correction
R2

.99

.93

.99

.99

A2

.98

.99
.99

.99
.99

SEE

.s4

0.05

0.04

0.03

0.01

dw

2.io

1.57

2.76

1.90

3 Period

2.20

1.56

4.37

135.9

Forecast F

* denotes significant at 1%. + at 5%; and ? at 10%. 1

-- ~
17

TABLE 3: ESTIMATED REVENUE EFFECTS OF AMNESTIES

(in mores of 1960 Rupees)
Model P Model U Amnesty

Year
1965

Receipts

Revenues

-119

-144

38

206

1966

-86

-103

210

1967

-84

-101

205

105

106

9#

438

1976

18

11

431

lY7.1

14

339

19k30*

-21

-22

43

433

-35

-28

357

1982

-34

-28

352

: 1983

-34

-25

345

1984

-35

-26

362

1985

-6

34

442

1986

-7

32

470

1987

-47

-34

478

1988

-57

-42

586

I 1989

-63

-66

650

1990

-61

-64

627

1991

-10

La.

691

171

-110

n.a.

1975

1981

1992

Total

'

734

*: Imputed rexreenue from Bearer Bond Scheme, allocated to 1980.

18

Upto

and including the 1980 amnesty. relevant amnesty and post-amnesty year effects

are statistically significant (except for the post-1975 period) and numerically sizeable. Indeed. the 1965 amnest>- is estimated to have reduced revenues by as much as 30-40%

during

amnesty and post-amnesty years. which appears implausibly large. The 1975 amnesty is
estimated on the other hand to have increased revenues by about 30% during the amnesty
year. with negligible after-effects.

The indirect effects of the 1980 amnesty during the

amnesty year are estimated to have reduced revenues by 15%, exceeding the direct yield
from the Bearer Bond Scheme - resuIting

in a net reduction of about 5% of annual revenues.

lrith the exception of 1975. therefore, indirect effects associated with amnesties served to
substantially reduce revenues even during the amnesty year. While this may be attributed
partly to induced mis-filing. it is plausible that the 1965-66 and 1980-81 amnesties:a]so
played a role in signal& c weak enforcement resolve of the government. The opposite was
the case with

the 1975-76 amnesty, as it coincided with the onset of the Emergency, causing

revenue increases in 1975-76

Folhnving

to exceed the drrect

receipts from the amnesty.

1980. when five-yearly amnesties came to be anticipated, interamnesty

years

are associated with a drop of the order of 10% a year. These possibly reflect intertemporal
substitution effects, as well as reduced collections from penarties
to declaration of black assets in previous amnesties. -3evertheless.
Iess significant during this period. hlorever.

and settlement fees owing

the revenue effects are

the 1985-87 and 1991-92 amnesties appear to

have had insignificant revenue effects during amnesty years. The negative indirect effects
during interamnesty years definiteI>.
directly

by

these

overshadowed whatever meagre revenues were yielded

amnesties.

One additional benefit. of an amnesty is that it may induce a reduction in the incidence of
non-filing behaviour:

by drawing more individuals into the tax net. The effect of the Indian

amnesties in increasing the total number of filers can be estimated by regressing the number
of income

tax assessees on amnesty dummies: after controlling for some basic features of

the tax structure and tax base (INFL. ?JAGDP.

AMTR and EXEAl).* The sample period

Fvas identical to that of the regressions reported above.

The four economic

variables were

significant and had the expected signs in both level and first difference (Engle-Grangerj
specificatiuns.

except

for EXEM Rhich

proved
19

to be insignificant in t.he first-dif&-xPnw

of the amnestv dummy variables were significant with the exception

specification.26 Uone
of the 1975 amnesty, v,-hich may simply be picking up,-an Emergency effect. Hence there is
no evidence that the amnesties served to reduce non-filing behaviour either.

5 Concluding Comments
The empirical results corroborate what one may expect from theory: that the indirect effects
of amnesties - arising

from

reduced

collections

from

penalties,

mis-filing

and

intertemporal

substitution - are adverse and can significantly overwhelm the positive direct effects. The
negative effects are less visible, being spread out both before and after the amnesty. Only
the 1975 amnesty appears to have exerted a significant positive effect on revenues: which
ma?- simply represent the effect of the state of Emergency that prevailed during the same
period. Al1 other amnesties exerted a negative overall effect. Morever,

amnesties appeared

to be increasingly insignificant after 1980. It is hard to avoid the conclusion that since 1980,
amncstv programs have run into negative or diminishing returns.

RIost amnesties appeared to have been sales, in the sense of offering a concession over
the tax normally levied on white income disclosures. This may have resulted in -mis-filing
of white incomes in the amnesty. causing additional revenue losses. Other forms of misfiling
include

disclosure of black assets

recently

discovered by tax authorities and di&oEure

in

the name of children, spouses and fictitious entities in order to lower the tax liability in
the anmestv . 27 The only possible justification for such low rates may have been to attract
habitual non-filers to participate in the amnesty. Nevertheless, there was no evidence to
suggest that amnesty programs served to draw non-filers into the tax net either. If the
government must persist with amnesty programs in the future, it should perhaps reconsider
Lvhether to continue to design them as sales.

20

References
Acharya. Shankar and associates (1985), -Aspects of th.e Black Economy in. India, National
Institute of Public Finance and Policy, New Delhi.
Allingham.

hl. and A. Sandmo (1972). %come Tax Evasion: A Theoretical Analysis,

Journal. of Public Economics. vol. 1, 322-338.

.41m. James and William Beck (1993). .Tax Amnesties and Compliance in the Long Run:
A Time Series Analysis. National Tax Journal, vol. XLVI(l), 53-60.
Andreoni. James (1991), Reasonable Doubt and the Optimal Magnitude of Fines: Should
the Penalty Fit the Crime.7, Rand Journal of Economics, vol. 22, 385-395.
Bagchi, Amaresh (1981): Special Bearer Bonds: A Retrospect,, Economic and Political
Neelzly: vol. 16. June.
Becker. G.S. (1968). C rime and Punishment: An Economic Approach. Journal-.of
Economy, vol. 76, 169-217.
Bolton. Patrick (1986)? .-The Principle of Rlaximum
ment of Economics! Harvard University.

Po&,tical

Deterrence Revisited. mimeo. Depart-

Comptroller and Auditor General (CAG) (1990): Report of the Comptroller and Auditor
General of In.dia, for the year ended 31 March 1989, Amnesty Scheme (1985), No. 7 of
1990, New Delhi: Government of India.
Das-Gupta Arindam. Radhika LahiLi ancl D. AJookhel;iee (1995), Yncorne Tax Compliance
in India: An Empirical Analysis. (revised) mimeo, New Brunswick: Rutgers University.
D. Mookheriee and D.P. Panta (1992), Income Tax Enforcement in India: A
Preliminar~Analysis..
Report, New Delhi: National Institute of Public Finance and Policy.
and D. Mookherjee
(1995),
ing Device. mimeo. Boston University.

A Theory of Tax Amnesties as an Asset Launder-

Dubin. Jeffrey. hI. Graetz and L. Wilde (1992), State Income Tax Amnesties: Causes!
Quartrrly Jovmal of Econ.om.ics. vol. 107(3). 1057-1070.
Malik. A.S. (1990) .4voidance.
Econ.omics, vol. 21. 341-353.

Screening and Optimum Enforcement, Rand Journal

of

&okher,jeeF D. (1989) *The Economics of Enforcement. mimeo. -Yew Delhi: Indian Statistical
Institute.
ican,

- (1992). .hlonitoring vis-a-vis Investigation in Enforcement of Law. ~merRevieu. vol. 82, 556-565.

Econom.ic

- (1994). .Marginal

Deterrence in Enforcement of Law,i Journal of Political

economy. 102(S), 1039-1066.

Polinskv,
llagnitude

hl. and S. Shave11 (1979), The Optimal Trade-Off Between the Probability and
of Fines, .dmerican. Econom.ir: Review, ~01. 69. 880-891.

21

(1993). *Enforcement Costs and the Optimal Magnitude and Probability of

Fines. Journal of ,hw* and Econmndcs. 33. 133-1.18.
Shaye&
0m.y.

S. (1991), %?pecific
vol. 99, 1088-1108.

versus General Enforcement of Law, Journ.al

of Political Econ,-

Stella, Peter (199(l), Do Tax Amnesties Work? APZIRC Bulletin., February 1990, 40-43.
(1991), An Economic Analysis of Tax Amnesties, Journal of Public
vol. 46, 383-400.

Economies,

Stigler, G.L. (1970), The Optimal Enforcement of Lawst Journal of Political Economz~,
vol. 78, 526-536.
Virmani Arvind (1986), Measuring Income Tax Incentives Under Evasion : Effective
lIargina and Average Rate Schedules, Discussion Paper Xo. DRD187, Development Research Department. World Bank.
Wilde. Louis (1992), Y.!riminal Choice. Sonmonetary Sanctions and Marginal Deterrence:
-4 Normatiye Analysis, International Review of Law an,d Economics. vol. 12(3), 333-344.

22

Notes
1. This paper forms part of a research project on reforms in Indian Income Tax Enl&cement. which was funded by a grant from the IRIS Center at the Eniversitv uf
Maryland. We are grateful to the National Institute of Public Finance and Policy,
Yew Delhi, for hosting this project. and to Dr. Amaresh Bagchi for his continued
support and inspiration. We also thank Arbind Modi, Radhika Lahiri, and Prabal
Ray Chaudhuri for advice and assistance at various stages of this project, and to Satu
Kahkonen for detailed comments on an earlier draft of this paper.
2. A large literature has subsequently explored modifications .of the Becker-AllinghamSandmo model to explain why optimal penalties may be non-maximal. Explanations
range from heterogeneity and risk-aversion of offenders and the likelihood of mistaken
convictions, to the infeasibility of graduating monitoring effort with the seriousness
of the offense. prosecution costs that rise with the penalty, and the impact of the
p e n a l t y level on the likelihood of conviction. See Andreoni (1991a) : Bolton (1986)?
h$alik (1990), Xlookherjee (1989). Xlookherjee and Png (1992, 1994), Pohnsky and
Shave11 (1979. 1993)! Shave11 (1991): Stigier (1970) and Wiide (1992).
3. See Das-Gupta, Alookherjee

and Panta

(1992) for further details.

4. The increase in revenues in 1975 may have reflected the effects of tightened enforcement during the internal Emergency taht was declared in the country, rather than the
effect of the amnesty per se: it is not possible to disentangle amnesty effects from the
impact of the Emergency itself.
5. India. has a number of legal provisions that resemble a permanent tm amnesty in many
respects. Examples are the Settlement Commission allowing for out-of-court settlement of penalties due and immunity from prosecution if past evasion is voluntarily
disclosed: explanation 5 to Section 271(l)(c) of the Income Tax Act for voluntary
disclosures during search and seizure operations; and Sec.tions
220(2A) and 273(l) of
the Income Tax Act providing respectively for waiver or reduction of interest, and a
once-in-a-lifetime amnesty from prosecution.
6 . For a review of literature concerning amnesties, see Das-Gupta and hiookherjee

(1995).

7. The latter permitted remittance in foreign exchange to any person in India with
immunity from all direct tax and exchange control regulations. Of the others, one
involved deposits with the National Housing Bank, one amended Section 273A of the
Income Tax Act to provide an additional opportunity to declare undisclosed income,
and Gold Bonds were put on sale with the sources of funds for their purchase being
granted immunity from investigation.
8. The 1985-86 amnesty was. however. introduced amidst the raid-raj of Finance Minister 1.P. Singh, and may therefore have signalled temporarily stronger enforcement
resolve.
9. Bagchis (1981) calculation indicates that the tax on income from these bonds was
effectively equal to the loss from the deviation below market interest rates. In other
words. the disclosures themselves invited no tax? though subsequent income from the
disclosed assets implicitly paid tax at the usual rate.
10. The CAG (1990) Report on the 1985-86 amnesty
scheme described the offer of blanket
I
waivers

of interest on past tax nhligatinns.

23

wnlch amnunted tn R nuhstanti~l.

rwwmw

loss of Rs. 4712 million (against an overall gain of Rs. 4588 million from income and
weaIth tax amnesty combined).
11. The discussion here abstracts from the possibility of a taxpayer resorting to a permanent amnesty when discovery of black assets by tax authorities is imminent.
1 2 . Additional advantages flow from reducing the size of the underground economy, if the
social rate of return is higher in the white than in the black sector.
13. See Da+Gupta, Mookherjee and Panta (1992) for evidence of neghgible prosecution
effort with respect to tax evasion offences, particularly in recent years. Discovery of
tax offenses are penalized more frequently by out-of-court settlement, which essentially
involve the payment of compnllnding fees or settlement fees.
14. See, for instance, the list of irregularities reported by CAG (1990) report on the
1985-86 Amnesty Scheme. It reveals that the rate of tax levied on amnesty income
disclosures amounted to only 13%!
13. The overall effect depends on the lags involved in levying penalties applicable to black
asset. discoveries in pre-amnesty years. In the absence of any lags, total coll&tions
will &se. while with sufficient delays in collecting penalties, they will fall.
16. This is based on the following assumptions: black assets do not earn a lower pre-tax
rate of return than white assets, the probability of tax authorities discovering black
assets is low and the aggregate stock of black assets is large relative to white income
switched to non- disclosure in anticipation of an amnesty.
17. For related empirical estimates of amnesty programs in various state governments in
the USA, SW Alm and Beck (1993) and Dubin, Graetz and Wilde (1992). The former
test the hypothesis that the 1985 Colorado amnesty had no effect on the level or trend
of tax collections, using time series data. The latter use a pooled cross-section time
series data set for different states to infer that amnesties did have a positive impact
on revenue growth rates.
18. This index is provided by the work disposal rate: i.e., the rate at which currently
pending assessments are disposed of. This rate has increased steadily in the last
fifteen or twenty .years. fnllowinE
the steep growth of workload of the income ta>c
department. relative to available manpower. The consequence has been a declining
quality of scrutiny assessments. The upward movement of the work disposal rate is
also negatively and strongly correlated with the decline in the fraction of taxpayers
audited under the scrutiny assessment scheme. It therefore proxies for both the quality
and frequency of income tax audits.
1 9 . The reader is referred to Das-Gupta, Lahiri and Mookherjee (1995) for details of the
definitivlr and construction of thcsc variables.
20. Nevertheless: the non-amnesty variables included in the regression by themselves explain over 90% of year-to-year variations in revenues.
21. 30 significant lagged variable influences have been detected, however, except those
captured in a dynamic error correction effect.
22. The homogeneity restrictions embodied in the parsimonious specification were not
rejected by t.he data at 5% significance level, whereas any further simplifying assumptions were rejected. In particular, the hypothesis that pre- and post-amnesty effects
lasted only one year. leas rejected.
24

the exception of the inflation rate. all non-amnest>

variables exhibit strong
time trends. In particular, they fail Dickey-Fuller stationarity tests. However, it is
well known that such tests are not very powerful for series as short as ours. The
results are therefore reported for both level regressions and first-difference regressions
including a dynamic error correction effect. The regressions happen to pass standard
co-integration tests.

23. Nith

24. These are calculated on the basis of the first difference regression coefficients.
25. Yote that NAGDP is the product of per capita nonagricultural income and total
population, so that the effects of both rising taxable income and rising population are
incorporated.
26. These results are not reported in the text. The R2 was 0.98 for the level regression
and 0.76 for the first difference regression.
27. Some corroborating information is provided by the GAG (1990) Report on the 1985436
amnestv scheme, which pointed out certain deficiencies in its framing andjmplementation. It drew attention to a number of cases of acceptance of disclosures despite
eva&n already having been detected by the income tax department. Tax on many
disclosures were also sought to be avoided by making them in the names of ladies.
minors and benamis and by declaring disproportionate income in one year.
28. The analysis of Virmani (1986) suggests
the tax evasion problem in India.

25

that thic forms a significant component

of