《Dvara Research & ILA：2024年FSP研究報告：評估印度2016年《破產和破產法》對借款人層面的影響（英文版）（32頁）.pdf》由會員分享，可在線閱讀，更多相關《Dvara Research & ILA：2024年FSP研究報告：評估印度2016年《破產和破產法》對借款人層面的影響（英文版）（32頁）.pdf（32頁珍藏版）》請在三個皮匠報告上搜索。

1、Assessing the Borrower-Level Impact of the Insolvency and Bankruptcy Code 2016:A STUDY OF THEFRESH STARTPROCESSAn Indian Ins?tute of Excellence in InsolvencyApril 2024In more than seven years,the Insolvency and Bankruptcy Code,2016,remains inoperative for all natural persons,except the guarantors of

2、 corporate debt.This essentially leaves most individuals,proprietorships(which constitute the majority of Indian micro,small,&medium enterprises,)and partnerships which are not limited liability entities,at the mercy of colonial legislations,which were products of the then political economy.As one m

3、ay expect for a country of nearly 1.5 billion,demographic heterogeneity is inevitable.On one hand,there are individuals higher up in the economic pyramid who can endure or often instigate long drawn and arduous litigations to optimize their relief under any insolvency and bankruptcy(hereafter bankru

4、ptcy)regime.On the other hand,there are low income households and enterprises who may need additional protections under the bankruptcy regime.Thus,looking at the present status,of the personal insolvency matters,there is huge opportunity for performing extensive research which in turn will help stre

5、ngthen Indias insolvency and bankruptcy framework for natural persons.Insolvency Law Academy and Dvara Research Foundation have jointly established a Chair for Personal Insolvency(Chair)to serve as a home for various research projects relating to personal insolvency,and micro,small and medium enterp

6、rises.The Chair focuses on the opportunity for research to help strengthen Indias insolvency and bankruptcy framework for individuals.We are pleased to present this paper written under the auspices of the Chair.Authored by Natasha Agnes Dcruze,Shree Harini V,Dwijaraj Bhattacharya,and Indradeep Ghosh

7、(all affiliated with Dvara Research),the paper was first presented at research conference of Centre for Advanced Financial Research and Learning(CAFRAL),an independent body set up by the Reserve Bank of India,held in Mumbai,India,in December,2023,and at ILA Annual Conference held in Goa,India,in Feb

8、ruary,2024.The paper has been finalized after considering feedback from both theconferences.Sumant Batra Indradeep GhoshPresidentInsolvency Law AcademyExecutive DirectorDvara Research FoundationDVARA RESEARCH FOUNDATIONDvara Research Foundation(Dvara)is a not-for-profit company policy research insti

9、tution based in India and incorporated under Section 25 of the Companies Act 1956.The main mission of Dvara is to ensure that every individual and every enterprise has complete access to financial services and occupy a unique position among the policy advocacy institutions in the country for researc

10、h and policy advocacy activities in financial services and financial inclusion.Since 2008,Dvara has made several contributions to the Indian financial system,specifically in the development of high-quality origination in India through research,and by assessing the impact of various financial product

11、s.Dvara has engaged with key policymaking institutions such as the Reserve Bank of India,Securities and Exchange Board of India,Pension Fund Regulatory and Development Authority,etc.,through its research INSOLVENCY LAW ACADEMYILA is an independent institute of excellence in insolvency.As a think tan

12、k and research institute,ILA seeks to contribute to robust policy making and enhancement of standards in the insolvency industry through cutting-edge research,innovation and development of best practices.ILA aims to develop a community in pursuit of education,research and scholarship in the field of

13、 insolvency in IABOUT In more than seven years,the Insolvency and Bankruptcy Code,2016,remains inoperative for all natural persons,except the guarantors of corporate debt.This essentially leaves most individuals,proprietorships(which constitute the majority of Indian micro,small,&medium enterprises,

14、)and partnerships which are not limited liability entities,at the mercy of colonial legislations,which were products of the then political economy.As one may expect for a country of nearly 1.5 billion,demographic heterogeneity is inevitable.On one hand,there are individuals higher up in the economic

15、 pyramid who can endure or often instigate long drawn and arduous litigations to optimize their relief under any insolvency and bankruptcy(hereafter bankruptcy)regime.On the other hand,there are low income households and enterprises who may need additional protections under the bankruptcy regime.Thu

16、s,looking at the present status,of the personal insolvency matters,there is huge opportunity for performing extensive research which in turn will help strengthen Indias insolvency and bankruptcy framework for natural persons.Insolvency Law Academy and Dvara Research Foundation have jointly establish

17、ed a Chair for Personal Insolvency(Chair)to serve as a home for various research projects relating to personal insolvency,and micro,small and medium enterprises.The Chair focuses on the opportunity for research to help strengthen Indias insolvency and bankruptcy framework for individuals.We are plea

18、sed to present this paper written under the auspices of the Chair.Authored by Natasha Agnes Dcruze,Shree Harini V,Dwijaraj Bhattacharya,and Indradeep Ghosh(all affiliated with Dvara Research),the paper was first presented at research conference of Centre for Advanced Financial Research and Learning(

19、CAFRAL),an independent body set up by the Reserve Bank of India,held in Mumbai,India,in December,2023,and at ILA Annual Conference held in Goa,India,in February,2024.The paper has been finalized after considering feedback from both theconferences.Sumant Batra Indradeep GhoshPresidentInsolvency Law A

20、cademyExecutive DirectorDvara Research FoundationDVARA RESEARCH FOUNDATIONDvara Research Foundation(Dvara)is a not-for-profit company policy research institution based in India and incorporated under Section 25 of the Companies Act 1956.The main mission of Dvara is to ensure that every individual an

21、d every enterprise has complete access to financial services and occupy a unique position among the policy advocacy institutions in the country for research and policy advocacy activities in financial services and financial inclusion.Since 2008,Dvara has made several contributions to the Indian fina

22、ncial system,specifically in the development of high-quality origination in India through research,and by assessing the impact of various financial products.Dvara has engaged with key policymaking institutions such as the Reserve Bank of India,Securities and Exchange Board of India,Pension Fund Regu

23、latory and Development Authority,etc.,through its research INSOLVENCY LAW ACADEMYILA is an independent institute of excellence in insolvency.As a think tank and research institute,ILA seeks to contribute to robust policy making and enhancement of standards in the insolvency industry through cutting-

24、edge research,innovation and development of best practices.ILA aims to develop a community in pursuit of education,research and scholarship in the field of insolvency in IABOUT The 2016 Insolvency and Bankruptcy Code(IBC)is a landmark legislation with the potential to impact every borrower.This pape

25、r focuses on Part III of the IBC,which deals with natural persons,proprietorships,and personal guarantors for corporate debt.Through the paper,we attempt to estimate the potential consequences of the Fresh Start Process(FSP)defined under this Part.The IBC lays out economic criteria that can qualify(

26、or disqualify)an applicant for FSP.Under FSP,a borrower must be asset-lite,have a low income,and hold minimal outstanding debt to qualify.These thresholds determine the applicability of the process once the IBC is fully notified.Thus,empirical estimates regarding the effects of the provisions on the

27、 Indian credit market are crucial to deciphering the impact of the IBC,more specifically,the FSP.We start by comparing the contemplated processes and outcomes of IBC with other similar legislations,like the Securitisation and Reconstruction of Financial Assets and Enforcement of Security Interest Ac

28、t(2002),Provincial Insolvency Act(1920),and Presidency Towns Insolvency Acts(1909).We then proceed to estimate how many borrowers are likely to qualify under the FSP.We use the Centre for Monitoring Indian Economys(CMIE)Consumer Pyramids Household Survey(CPHS)conjoined(using a nearest neighbour mode

29、l and the Hungarian Algorithm)with the All-India Debts and Investments Survey(AIDIS)for 2019 to estimate how many households qualify under FSP.We perform the analysis for the entire country,except a few states and union territories with relatively sparse population.Thus,our research is intended as a

30、 methodological contribution through which the impact of the IBC across borrower groups can be measured.ABSTRACT:The 2016 Insolvency and Bankruptcy Code(IBC)is a landmark legislation with the potential to impact every borrower.This paper focuses on Part III of the IBC,which deals with natural person

31、s,proprietorships,and personal guarantors for corporate debt.Through the paper,we attempt to estimate the potential consequences of the Fresh Start Process(FSP)defined under this Part.The IBC lays out economic criteria that can qualify(or disqualify)an applicant for FSP.Under FSP,a borrower must be

32、asset-lite,have a low income,and hold minimal outstanding debt to qualify.These thresholds determine the applicability of the process once the IBC is fully notified.Thus,empirical estimates regarding the effects of the provisions on the Indian credit market are crucial to deciphering the impact of t

33、he IBC,more specifically,the FSP.We start by comparing the contemplated processes and outcomes of IBC with other similar legislations,like the Securitisation and Reconstruction of Financial Assets and Enforcement of Security Interest Act(2002),Provincial Insolvency Act(1920),and Presidency Towns Ins

34、olvency Acts(1909).We then proceed to estimate how many borrowers are likely to qualify under the FSP.We use the Centre for Monitoring Indian Economys(CMIE)Consumer Pyramids Household Survey(CPHS)conjoined(using a nearest neighbour model and the Hungarian Algorithm)with the All-India Debts and Inves

35、tments Survey(AIDIS)for 2019 to estimate how many households qualify under FSP.We perform the analysis for the entire country,except a few states and union territories with relatively sparse population.Thus,our research is intended as a methodological contribution through which the impact of the IBC

36、 across borrower groups can be measured.ABSTRACT:The Insolvency and Bankruptcy Code,2016(IBC)was introduced in an environment where formal sector lenders,especially banks,struggled with low asset quality.The IBC was intended“to consolidate and amend the laws relating to reorganisation and insolvency

37、 resolution of corporate persons,partnership firms and individuals in a time bound manner for maximisation of value of assets of such persons,to promote entrepreneurship,availability of credit and balance the interests of all the 1stakeholders”.It has been almost 8 years,but not all stakeholders are

38、 still covered.Currently,the act is operational(i.e.,notified by the government)for corporate debtors and individuals(natural persons)who are guarantors of corporate debtors.Non-limited liability entities like partnerships,proprietorships,etc.,are still outside the scope of the remedies proposed by

39、the IBC,since Part III of the IBC,which deals with such debtors,is not notified in its entirety.Though there is no official declaration regarding why some sections of the code are not notified,it is possible to conjecture that they have to do with the rather complex issue of natural persons.In the c

40、ase of Part III of the IBC,a human subject in distress becomes a key consideration.Policymakers must,therefore,contend not only with how Part III will impact credit markets but also with the ethical question of whether a natural person deserves relief in some form and,if so,why.The three processes o

41、utlined in the IBC provide us a glimpse into the minds of the policymakers,especially highlighting how they envision answering this ethical question.The three processes under which a natural person(or her creditor)may seek shelter are:a)The Insolvency Resolution Process(IRP)b)The Bankruptcy Process,

42、and c)The Fresh Start Process(FSP).The first two processes form part of a continuum,whereby any debtor(or their creditor)can file for an IRP and apply for bankruptcy if such an IRP fails.The third process,FSP,is unique.It is targeted towards low-income borrowers who are asset-light and have minimal

43、outstanding debt,i.e.,the most vulnerable borrowers.2For such qualifying individuals,the FSP proposes a scenario where their debts can be wiped clean,i.e.,“discharged”.In this paper,we situate the FSP in the historical arc of insolvency and bankruptcy regimes and processes,and then present a methodo

44、logy(and insights therefrom)through which the impact of the FSP can be measured at a borrower level.I.Introduction1 Per the long title of the IBC,20162 Specific qualification criteria are discussed in the next section.0102For as long as credit has existed,there have been borrowers unable to repay th

45、eir monetary debts,and an attempt to recover the debts has always led to acrimony.In classical antiquity,creditors could repossess the debtors person,i.e.,debt slavery was common,and the practice was rooted in customs rather than formal laws(Levinthal,1918).Between stththe 1 and 16 century AD,a seco

46、nd phase of insolvency practices developed;debt slavery received formal legal sanction,but certain sections of the society(members of higher political standing)were granted immunity from such a punishment.With the dawn of enlightenment,rational-legal principles thbegan to take centre stage,and by th

47、e mid-16 century,formal law offered some protection to the debtor in default but also empowered the state(more precisely,its embodiment,the crown)to impose the death penalty(Carlos,2019;Bhattacharya&Ghosh,2022)Across these three phases,the purpose of the law(or the custom)was to enable the creditor

48、to reclaim their debt.Further,another common feature unites these three phases-the lenders and borrowers were 3mostly singular entities and natural persons.However,there were exceptions to this rule,i.e.,some institutions did lend and borrow.After the 10th century,institutions like the church and th

49、e crown(s)often received or disbursed credit.The terms of such credit were,however,governed by bilateral agreements between the lender and the borrower rather than a codified national-level law.Starting from the turn of the 19th century,the modern era marks a significant departure from the earlier p

50、hases.Natural persons no longer occupy centre stage,neither as 45creditors nor as borrowers.With the invention of II.Situating the Fresh Start Process in a Historical Contextcompanies,and with such companies receiving the lions share of credit(earlier for trade and manufacturing,and later for servic

51、es),they emerge as the key focus for insolvency and bankruptcy regimes(Bhattacharya&Ghosh,2022).Axiomatically,we know that corporations are different from natural persons.The former can be carved into pieces and liquidated.The latter,on the other hand,have inalienable rights.Therefore,modern-day ins

52、olvency and bankruptcy regimes have attempted to move beyond the express purpose of enabling creditors to reclaim their debt.Now,they aim to balance the rights of the debtors against the creditors.In India,the Presidency Towns Insolvency Act(1909)and the Provincial Insolvency Act(1920)attempted to d

53、o this before Part III of the IBC sought to replace them with new provisions.However,those earlier acts remain in force since the majority of Part 6III of the IBC is yet to be notified.Apart from procedural aspects such as the identification of the forum having jurisdiction over the subject,the pres

54、ence of a moratorium,the time-bound nature of the processes,the need for an insolvency resolution professional etc.,the key difference between the British-era statutes and the IBC is the Fresh Start Process(FSP).The FSP is a low-cost quasi-bankruptcy process applicable for low-income,asset-light deb

55、tors holding minimal debt.It allows for a complete discharge of their debt provided they satisfy specific economic and procedural criteria.Thus,the FSP process mimics the gate-kept bankruptcy process whereby the debtor may get a complete discharge from their obligations(Bhattacharya&Ananth,2021).3 o

56、ne person would be lending to another rather than a consortium of persons lending to one or more people 4 replaced by banks,and banking institutions5 replaced by corporates 6 since section(243)of the IBC which repeals the Presidency Towns Insolvency Act(1909)and the Provincial Insolvency Act(1920)ha

57、s not been notified,these laws remain in-force.The Insolvency and Bankruptcy Code,2016(IBC)was introduced in an environment where formal sector lenders,especially banks,struggled with low asset quality.The IBC was intended“to consolidate and amend the laws relating to reorganisation and insolvency r

58、esolution of corporate persons,partnership firms and individuals in a time bound manner for maximisation of value of assets of such persons,to promote entrepreneurship,availability of credit and balance the interests of all the 1stakeholders”.It has been almost 8 years,but not all stakeholders are s

59、till covered.Currently,the act is operational(i.e.,notified by the government)for corporate debtors and individuals(natural persons)who are guarantors of corporate debtors.Non-limited liability entities like partnerships,proprietorships,etc.,are still outside the scope of the remedies proposed by th

60、e IBC,since Part III of the IBC,which deals with such debtors,is not notified in its entirety.Though there is no official declaration regarding why some sections of the code are not notified,it is possible to conjecture that they have to do with the rather complex issue of natural persons.In the cas

61、e of Part III of the IBC,a human subject in distress becomes a key consideration.Policymakers must,therefore,contend not only with how Part III will impact credit markets but also with the ethical question of whether a natural person deserves relief in some form and,if so,why.The three processes out

62、lined in the IBC provide us a glimpse into the minds of the policymakers,especially highlighting how they envision answering this ethical question.The three processes under which a natural person(or her creditor)may seek shelter are:a)The Insolvency Resolution Process(IRP)b)The Bankruptcy Process,an

63、d c)The Fresh Start Process(FSP).The first two processes form part of a continuum,whereby any debtor(or their creditor)can file for an IRP and apply for bankruptcy if such an IRP fails.The third process,FSP,is unique.It is targeted towards low-income borrowers who are asset-light and have minimal ou

64、tstanding debt,i.e.,the most vulnerable borrowers.2For such qualifying individuals,the FSP proposes a scenario where their debts can be wiped clean,i.e.,“discharged”.In this paper,we situate the FSP in the historical arc of insolvency and bankruptcy regimes and processes,and then present a methodolo

65、gy(and insights therefrom)through which the impact of the FSP can be measured at a borrower level.I.Introduction1 Per the long title of the IBC,20162 Specific qualification criteria are discussed in the next section.0102For as long as credit has existed,there have been borrowers unable to repay thei

66、r monetary debts,and an attempt to recover the debts has always led to acrimony.In classical antiquity,creditors could repossess the debtors person,i.e.,debt slavery was common,and the practice was rooted in customs rather than formal laws(Levinthal,1918).Between stththe 1 and 16 century AD,a second

67、 phase of insolvency practices developed;debt slavery received formal legal sanction,but certain sections of the society(members of higher political standing)were granted immunity from such a punishment.With the dawn of enlightenment,rational-legal principles thbegan to take centre stage,and by the

68、mid-16 century,formal law offered some protection to the debtor in default but also empowered the state(more precisely,its embodiment,the crown)to impose the death penalty(Carlos,2019;Bhattacharya&Ghosh,2022)Across these three phases,the purpose of the law(or the custom)was to enable the creditor to

69、 reclaim their debt.Further,another common feature unites these three phases-the lenders and borrowers were 3mostly singular entities and natural persons.However,there were exceptions to this rule,i.e.,some institutions did lend and borrow.After the 10th century,institutions like the church and the

70、crown(s)often received or disbursed credit.The terms of such credit were,however,governed by bilateral agreements between the lender and the borrower rather than a codified national-level law.Starting from the turn of the 19th century,the modern era marks a significant departure from the earlier pha

71、ses.Natural persons no longer occupy centre stage,neither as 45creditors nor as borrowers.With the invention of II.Situating the Fresh Start Process in a Historical Contextcompanies,and with such companies receiving the lions share of credit(earlier for trade and manufacturing,and later for services

72、),they emerge as the key focus for insolvency and bankruptcy regimes(Bhattacharya&Ghosh,2022).Axiomatically,we know that corporations are different from natural persons.The former can be carved into pieces and liquidated.The latter,on the other hand,have inalienable rights.Therefore,modern-day insol

73、vency and bankruptcy regimes have attempted to move beyond the express purpose of enabling creditors to reclaim their debt.Now,they aim to balance the rights of the debtors against the creditors.In India,the Presidency Towns Insolvency Act(1909)and the Provincial Insolvency Act(1920)attempted to do

74、this before Part III of the IBC sought to replace them with new provisions.However,those earlier acts remain in force since the majority of Part 6III of the IBC is yet to be notified.Apart from procedural aspects such as the identification of the forum having jurisdiction over the subject,the presen

75、ce of a moratorium,the time-bound nature of the processes,the need for an insolvency resolution professional etc.,the key difference between the British-era statutes and the IBC is the Fresh Start Process(FSP).The FSP is a low-cost quasi-bankruptcy process applicable for low-income,asset-light debto

76、rs holding minimal debt.It allows for a complete discharge of their debt provided they satisfy specific economic and procedural criteria.Thus,the FSP process mimics the gate-kept bankruptcy process whereby the debtor may get a complete discharge from their obligations(Bhattacharya&Ananth,2021).3 one

77、 person would be lending to another rather than a consortium of persons lending to one or more people 4 replaced by banks,and banking institutions5 replaced by corporates 6 since section(243)of the IBC which repeals the Presidency Towns Insolvency Act(1909)and the Provincial Insolvency Act(1920)has

78、not been notified,these laws remain in-force.In the present form,an individual(debtor)applying for FSP under the IBC must satisfy four economic criteria,as specified in sections 80(2)(a)-80(2)(c)and 80(2)(e)of the IBC.These include the income criterion(the debtor must have annual income not exceedin

79、g 60,000),the asset criterion(the aggregate value of the debtors assets ought not to exceed 20,000),the debt criterion(the eligible debt owed by the individual must not exceed 35,000)and an extension of the asset criterion,whereunder for a debtor to be eligible,they must not own a dwelling unit.Furt

80、her,the IBC specifies that these criteria should be jointly applied,meaning that a debtor would qualify for the FSP if it satisfies all four(IBC,2016).The criteria,however,leave significant scope for interpretation in their definitions.For instance,it is unclear which income streams would be conside

81、red income under the income criterion.For an individual operating a proprietorship,all revenues from the business venture are essentially personal income and that aggregate number is very likely to exceed the ceiling,thus making most ineligible for the remedy.Furthermore,it is unclear whether direct

82、 benefits transfers by the government will be considered income.If they are,that would even further reduce the eligible debtor numbers.The asset criterion and its extension present several dilemmas also.How do we ascertain the value of household goods?Who should be considered the owner if the asset

83、is a common asset?Regarding the ownership of a dwelling,how should structures that are not wholly residential but used for residential purposes be treated(e.g.a hut on agricultural land used as the residence and storage unit for grains)?Thus,estimating the impact of the IBC,especially the FSP,using

84、an as-is interpretation of Part III must be accompanied by a set of assumptions that seek to resolve interpretive concerns such as the ones identified in the previous paragraph.The following section discusses these assumptions and the data sources(and their transformations)in detail.0304In India,no

85、pan-national official data source simultaneously captures an individuals income,the assets owned by them and their debts.These data reside in fragmented silos.For income,the official data resides within the income tax department.However,with only 7.4 crore people filing income tax returns in 2022-23

86、 and given the widespread informal economy in the country,the data is neither comprehensive nor adequately representative.For data on debt owed by the individual,the hurdles are similar.Credit Information Companies(CICs)capture the cumulative credit outstanding for individuals and businesses,but the

87、 data only represents formal credit,thus reducing representativeness and comprehensiveness.Most importantly,however,neither of the above data sources is public.Finally,capturing the asset ownership of an individual through any consolidated database is virtually impossible.So,official data sources ar

88、e of little help,and reliance must be placed on nationally representative surveys for estimation efforts.Currently,two such surveys exist the All-India Debt and Investment Survey(AIDIS),conducted by the National Sample Survey Organisation(NSSO)and the Consumer Pyramids Household Surveys(CPHS),conduc

89、ted by the Center for Monitoring Indian Economy(CMIE).Both surveys have their limitations.AIDIS is a sample survey that captures quantitative information on assets and liabilities but not income.Further,most of the relevant data for our analysis is captured at a household level and not at the indivi

90、dual level,which ought to be the unit of analysis given the construct of the FSP.The CPHS,on the other hand,provides complementary details,like the quantum of income and ownership of debt,at an individual level.And also,across most asset segments such as household durables,jewellery,vehicles,etc.,th

91、e CPHS data only indicates whether a particular asset types is owned or not,and not its value(if owned).Thus,from CPHS,we may only learn that a household has jewellery,but not how much it is worth.The qualification criteria for FSP,however,are based on values.III.Data Sources and Methods Thus,neithe

92、r the CPHS dataset nor the AIDIS dataset can be used in isolation to estimate the number of borrowers the FSP will cover.However,together,both datasets complement each other.The CPHS dataset presents select insights at an individual level and captures income.In contrast,the AIDIS dataset captures gr

93、anular details on asset ownership and debt owed,though at a household level.Thus,a combined analysis of both datasets is critical,necessitating us to adopt an approach to match households from one dataset to another.Section 3.1:Matching the Datasets Matching observations between datasets is a common

94、 yet intricate challenge,especially when dealing with sample surveys representing the same universe.This task becomes particularly complex since the AIDIS(for the year 2019)and CPHS(for the year 2019)datasets have a multitude of variables,both categorical and continuous.These variables must be taken

95、 into account simultaneously for any accurate matching.This objective can,therefore,be recast as a classification problem.To elucidate,let us consider there are three households,a,a,and a 123from the AIDIS dataset and c,c and c from the 123CPHS dataset.Further,let us consider there are three variabl

96、es common between the two datasets,V,1V,and V.The values of each variable for the 23different households are given below in Tables 1(A)and 1(B).Householda1a2a3V1MaleMaleFemaleV2756V3150003000045000Table-1(A):Snippet from AIDIS dataset Householdc1c2c3V1MaleFemaleFemaleV27126V31500015000048000Table-1(

97、B):Snippet from CPHS dataset In the present form,an individual(debtor)applying for FSP under the IBC must satisfy four economic criteria,as specified in sections 80(2)(a)-80(2)(c)and 80(2)(e)of the IBC.These include the income criterion(the debtor must have annual income not exceeding 60,000),the as

98、set criterion(the aggregate value of the debtors assets ought not to exceed 20,000),the debt criterion(the eligible debt owed by the individual must not exceed 35,000)and an extension of the asset criterion,whereunder for a debtor to be eligible,they must not own a dwelling unit.Further,the IBC spec

99、ifies that these criteria should be jointly applied,meaning that a debtor would qualify for the FSP if it satisfies all four(IBC,2016).The criteria,however,leave significant scope for interpretation in their definitions.For instance,it is unclear which income streams would be considered income under

100、 the income criterion.For an individual operating a proprietorship,all revenues from the business venture are essentially personal income and that aggregate number is very likely to exceed the ceiling,thus making most ineligible for the remedy.Furthermore,it is unclear whether direct benefits transf

101、ers by the government will be considered income.If they are,that would even further reduce the eligible debtor numbers.The asset criterion and its extension present several dilemmas also.How do we ascertain the value of household goods?Who should be considered the owner if the asset is a common asse

102、t?Regarding the ownership of a dwelling,how should structures that are not wholly residential but used for residential purposes be treated(e.g.a hut on agricultural land used as the residence and storage unit for grains)?Thus,estimating the impact of the IBC,especially the FSP,using an as-is interpr

103、etation of Part III must be accompanied by a set of assumptions that seek to resolve interpretive concerns such as the ones identified in the previous paragraph.The following section discusses these assumptions and the data sources(and their transformations)in detail.0304In India,no pan-national off

104、icial data source simultaneously captures an individuals income,the assets owned by them and their debts.These data reside in fragmented silos.For income,the official data resides within the income tax department.However,with only 7.4 crore people filing income tax returns in 2022-23 and given the w

105、idespread informal economy in the country,the data is neither comprehensive nor adequately representative.For data on debt owed by the individual,the hurdles are similar.Credit Information Companies(CICs)capture the cumulative credit outstanding for individuals and businesses,but the data only repre

106、sents formal credit,thus reducing representativeness and comprehensiveness.Most importantly,however,neither of the above data sources is public.Finally,capturing the asset ownership of an individual through any consolidated database is virtually impossible.So,official data sources are of little help

107、,and reliance must be placed on nationally representative surveys for estimation efforts.Currently,two such surveys exist the All-India Debt and Investment Survey(AIDIS),conducted by the National Sample Survey Organisation(NSSO)and the Consumer Pyramids Household Surveys(CPHS),conducted by the Cente

108、r for Monitoring Indian Economy(CMIE).Both surveys have their limitations.AIDIS is a sample survey that captures quantitative information on assets and liabilities but not income.Further,most of the relevant data for our analysis is captured at a household level and not at the individual level,which

109、 ought to be the unit of analysis given the construct of the FSP.The CPHS,on the other hand,provides complementary details,like the quantum of income and ownership of debt,at an individual level.And also,across most asset segments such as household durables,jewellery,vehicles,etc.,the CPHS data only

110、 indicates whether a particular asset types is owned or not,and not its value(if owned).Thus,from CPHS,we may only learn that a household has jewellery,but not how much it is worth.The qualification criteria for FSP,however,are based on values.III.Data Sources and Methods Thus,neither the CPHS datas

111、et nor the AIDIS dataset can be used in isolation to estimate the number of borrowers the FSP will cover.However,together,both datasets complement each other.The CPHS dataset presents select insights at an individual level and captures income.In contrast,the AIDIS dataset captures granular details o

112、n asset ownership and debt owed,though at a household level.Thus,a combined analysis of both datasets is critical,necessitating us to adopt an approach to match households from one dataset to another.Section 3.1:Matching the Datasets Matching observations between datasets is a common yet intricate c

113、hallenge,especially when dealing with sample surveys representing the same universe.This task becomes particularly complex since the AIDIS(for the year 2019)and CPHS(for the year 2019)datasets have a multitude of variables,both categorical and continuous.These variables must be taken into account si

114、multaneously for any accurate matching.This objective can,therefore,be recast as a classification problem.To elucidate,let us consider there are three households,a,a,and a 123from the AIDIS dataset and c,c and c from the 123CPHS dataset.Further,let us consider there are three variables common betwee

115、n the two datasets,V,1V,and V.The values of each variable for the 23different households are given below in Tables 1(A)and 1(B).Householda1a2a3V1MaleMaleFemaleV2756V3150003000045000Table-1(A):Snippet from AIDIS dataset Householdc1c2c3V1MaleFemaleFemaleV27126V31500015000048000Table-1(B):Snippet from

116、CPHS dataset Datasets like the AIDIS and CPHS often contain variables like the gender of the head of the household,the number of members in the family and the income(at a given frequency).Thus,we can assume V,V,and V represent these categories.123With the presented information,it would appear that h

117、ouseholds a and c are identical since any and all 11given variables have identical values.Conversely,households a and c are very different.In the case of 22a and c,however,concluding whether the 33households are identical(or different)is an arduous task,especially when we consider that data may have

118、 been collected at different points in time.Thus,statistical models must be used to systematically calculate similarities between two households using their properties(i.e.,variables).One of the most popular methods for solving such classification problems is the k-nearest-neighbour(KNN)method(Cover

119、&Hart,1967).The KNN method is often used for classification and regression tasks(Fix&Hodges Jr,1951).Its flexibility and simplicity make it a valuable tool in data-matching exercises.The method operates on the premise that similar instances in the feature(i.e.,variable)space tend to share similar la

120、bels(Song et al.,2017).In the context of our exercise,it means that households with similar characteristics,such as the number of members,location,social group,expenditure,etc.,are likely to be the same,i.e.,they reflect identical characteristics.Thus,the KNN method essentially establishes similarit

121、ies(Mehta et al.,2018),which can then be inferred to mean that household a from AIDIS is identical to household c from CPHS.Before proceeding further,it is important to understand the KNN methods three key aspects.First,how is the distance between the neighbours calculated?Second,how is the value of

122、 K assigned?Third,how is the assignment decision made(decision rule)?On the choice of distance measure,we note first that we are working with two types of variables:categorical ones and continuous ones.A categorical variable can assume a finite number of categories without a natural ordering.For exa

123、mple,the states of India may be coded as numbers,with 1 representing Andhra Pradesh,2 for Arunachal Pradesh,28 representing West Bengal,and so on(assignment per alphabetical order).Here,the numbers 1 to 28 have a natural order,where 28 is greater than 27,which in turn is greater than 26,and so on.Ho

124、wever,such ordering is meaningless.Just because West Bengal is 28 and Andhra Pradesh is 1,it doesnt mean West Bengal is greater than Andhra Pradesh.Similarly,in our case,the categorical variables discussed in Table-2 do not share a natural order,despite often being coded as numbers.The second variab

125、le type is a quantitative measurement(on the integers or real numbers line).In this case,there is a natural order.Further,the difference between the values are also meaningful.For example,an expense of 10 is less than one of 100.Similarly,the difference between 10 and 100 is meaningful since we can

126、now learn that one household consumed more goods valued and we can quantify that difference as 90 in value terms.Variable Description Region(Urban/Rural)District Social Group Religion Age Groups Gender Groups Household Size Groups Household Expenditure#Similar HHs in the Country Variable Type Catego

127、rical Categorical Categorical Categorical Categorical Categorical Categorical Continuous Continuous Table-2:Variables selected for identifyingsimilar households Source:Authors Calculations 0506Several distance functions are available when dealing with all categorical or non-categorical variables(Abu

128、 Alfeilat et al.,2019;Van de Velden et al.,2019).However,options are limited for datasets with mixed-type variables,which is common in survey data.The continuous variables are normalised first so that the values lie between 0 and 1.The normalisation is achieved by subtracting the minimum value of th

129、e variable in the dataset from the value to be normalised and dividing this difference by the difference between the maximum and minimum values of the variable in the dataset.Thereafter,we compute the scalar distance for the normalised variable between the two households(from the AIDIS and CPHS data

130、sets).Thus,we obtain 2 distances,one for each variable.To resolve these 2 distances into a single measure that combines the distance for all(both)continuous variables,we square each scalar difference,then sum the squares and then take the square root(this is a Euclidean metric).This result is divide

131、d by 2 to obtain a continuous distance distribution(between 0 and 1).For categorical variables,the process is more straightforward.For each of the categorical variables,either there will be a perfect match or not.If there is a perfect match,we calculate that distance as zero.If not,then we calculate

132、 that distance as 1.We then sum the 7 distances(for the seven categorical variables)to obtain a combined measure of the distance for all categorical variables.This result is 7divided by 7 to obtain a step-separated categorical distance(between 0 and 1).Finally,the two distance measures,one for conti

133、nuous variables and the other for categorical variables,are resolved into a single distance measure using the modified Gower method(Gower,1971),and this too produces a number between 0 and 1.This concludes the discussion on the first of the three aspects of the KNN method.The second and third aspect

134、s are the value assigned to K and the assignment algorithm for the nearest match.We discuss these together as they relate closely to each other.For our estimation,we assign the value of 5 to K,meaning that the KNN method will consider the 5 nearest neighbours(based on the collapsed distance as measu

135、red through the modified Gowers distance)before assigning which is the closest match(based on the individual distances across all variables).For our analysis,we can consider that the operation is being carried out for household a from AIDIS across all households c 1to c from CMIE.Thus,in the first s

136、tep,the KNN nmethod will select 5 closest neighbours from the CPHS dataset using only one distance measure,the modified Gower distance.Thus,we obtain 5 possible assignments:household a matched to c (denoted 1as c a),or c a,c a,c a,and c5 a.In 1234a scenario where only one pair has the minimum distan

137、ce between two households,such a pair is considered to be the final match.To exemplify,if the distance between c a is 0.1 and the distances 1between c a,c a,etc.are all greater than 0.1,23household c is assigned to household a.However,1if the minimum distance is shared by two(or more)pairs,i.e.,the

138、distance between,say,c a and c 12a are identical and the minimum,then there is a tie.In such a scenario,to resolve the tie,the model computes 9 measures of distance for each pair of households,i.e.,for the pair(c a),the model 1computes the distance using the region variable,then the district variabl

139、e,and so on,across all variables listed in Table-2.So,instead of comparing just one distance measure,the model now compares nine distance measures to find which pair has the maximum number of minimum distances.It is still theoretically possible not to be able to resolve the tie;however,since we did

140、not face the situation,a discussion of the same is avoided.Through this process,the KNN method chooses which of the five households from CMIE is the closest match to household a of AIDIS.7 The distances are step separated,since it can only assume discrete values of 0/7(i.e.,all the categorical varia

141、bles match),or 1/7(i.e.,only one categorical variable does not match),and so on.Datasets like the AIDIS and CPHS often contain variables like the gender of the head of the household,the number of members in the family and the income(at a given frequency).Thus,we can assume V,V,and V represent these

142、categories.123With the presented information,it would appear that households a and c are identical since any and all 11given variables have identical values.Conversely,households a and c are very different.In the case of 22a and c,however,concluding whether the 33households are identical(or differen

143、t)is an arduous task,especially when we consider that data may have been collected at different points in time.Thus,statistical models must be used to systematically calculate similarities between two households using their properties(i.e.,variables).One of the most popular methods for solving such

144、classification problems is the k-nearest-neighbour(KNN)method(Cover&Hart,1967).The KNN method is often used for classification and regression tasks(Fix&Hodges Jr,1951).Its flexibility and simplicity make it a valuable tool in data-matching exercises.The method operates on the premise that similar in

145、stances in the feature(i.e.,variable)space tend to share similar labels(Song et al.,2017).In the context of our exercise,it means that households with similar characteristics,such as the number of members,location,social group,expenditure,etc.,are likely to be the same,i.e.,they reflect identical ch

146、aracteristics.Thus,the KNN method essentially establishes similarities(Mehta et al.,2018),which can then be inferred to mean that household a from AIDIS is identical to household c from CPHS.Before proceeding further,it is important to understand the KNN methods three key aspects.First,how is the di

147、stance between the neighbours calculated?Second,how is the value of K assigned?Third,how is the assignment decision made(decision rule)?On the choice of distance measure,we note first that we are working with two types of variables:categorical ones and continuous ones.A categorical variable can assu

148、me a finite number of categories without a natural ordering.For example,the states of India may be coded as numbers,with 1 representing Andhra Pradesh,2 for Arunachal Pradesh,28 representing West Bengal,and so on(assignment per alphabetical order).Here,the numbers 1 to 28 have a natural order,where

149、28 is greater than 27,which in turn is greater than 26,and so on.However,such ordering is meaningless.Just because West Bengal is 28 and Andhra Pradesh is 1,it doesnt mean West Bengal is greater than Andhra Pradesh.Similarly,in our case,the categorical variables discussed in Table-2 do not share a n

150、atural order,despite often being coded as numbers.The second variable type is a quantitative measurement(on the integers or real numbers line).In this case,there is a natural order.Further,the difference between the values are also meaningful.For example,an expense of 10 is less than one of 100.Simi

151、larly,the difference between 10 and 100 is meaningful since we can now learn that one household consumed more goods valued and we can quantify that difference as 90 in value terms.Variable Description Region(Urban/Rural)District Social Group Religion Age Groups Gender Groups Household Size Groups Ho

152、usehold Expenditure#Similar HHs in the Country Variable Type Categorical Categorical Categorical Categorical Categorical Categorical Categorical Continuous Continuous Table-2:Variables selected for identifyingsimilar households Source:Authors Calculations 0506Several distance functions are available

153、 when dealing with all categorical or non-categorical variables(Abu Alfeilat et al.,2019;Van de Velden et al.,2019).However,options are limited for datasets with mixed-type variables,which is common in survey data.The continuous variables are normalised first so that the values lie between 0 and 1.T

154、he normalisation is achieved by subtracting the minimum value of the variable in the dataset from the value to be normalised and dividing this difference by the difference between the maximum and minimum values of the variable in the dataset.Thereafter,we compute the scalar distance for the normalis

155、ed variable between the two households(from the AIDIS and CPHS datasets).Thus,we obtain 2 distances,one for each variable.To resolve these 2 distances into a single measure that combines the distance for all(both)continuous variables,we square each scalar difference,then sum the squares and then tak

156、e the square root(this is a Euclidean metric).This result is divided by 2 to obtain a continuous distance distribution(between 0 and 1).For categorical variables,the process is more straightforward.For each of the categorical variables,either there will be a perfect match or not.If there is a perfec

157、t match,we calculate that distance as zero.If not,then we calculate that distance as 1.We then sum the 7 distances(for the seven categorical variables)to obtain a combined measure of the distance for all categorical variables.This result is 7divided by 7 to obtain a step-separated categorical distan

158、ce(between 0 and 1).Finally,the two distance measures,one for continuous variables and the other for categorical variables,are resolved into a single distance measure using the modified Gower method(Gower,1971),and this too produces a number between 0 and 1.This concludes the discussion on the first

159、 of the three aspects of the KNN method.The second and third aspects are the value assigned to K and the assignment algorithm for the nearest match.We discuss these together as they relate closely to each other.For our estimation,we assign the value of 5 to K,meaning that the KNN method will conside

160、r the 5 nearest neighbours(based on the collapsed distance as measured through the modified Gowers distance)before assigning which is the closest match(based on the individual distances across all variables).For our analysis,we can consider that the operation is being carried out for household a fro

161、m AIDIS across all households c 1to c from CMIE.Thus,in the first step,the KNN nmethod will select 5 closest neighbours from the CPHS dataset using only one distance measure,the modified Gower distance.Thus,we obtain 5 possible assignments:household a matched to c (denoted 1as c a),or c a,c a,c a,an

162、d c5 a.In 1234a scenario where only one pair has the minimum distance between two households,such a pair is considered to be the final match.To exemplify,if the distance between c a is 0.1 and the distances 1between c a,c a,etc.are all greater than 0.1,23household c is assigned to household a.Howeve

163、r,1if the minimum distance is shared by two(or more)pairs,i.e.,the distance between,say,c a and c 12a are identical and the minimum,then there is a tie.In such a scenario,to resolve the tie,the model computes 9 measures of distance for each pair of households,i.e.,for the pair(c a),the model 1comput

164、es the distance using the region variable,then the district variable,and so on,across all variables listed in Table-2.So,instead of comparing just one distance measure,the model now compares nine distance measures to find which pair has the maximum number of minimum distances.It is still theoretical

165、ly possible not to be able to resolve the tie;however,since we did not face the situation,a discussion of the same is avoided.Through this process,the KNN method chooses which of the five households from CMIE is the closest match to household a of AIDIS.7 The distances are step separated,since it ca

166、n only assume discrete values of 0/7(i.e.,all the categorical variables match),or 1/7(i.e.,only one categorical variable does not match),and so on.households from the CPHS dataset to households in the AIDIS dataset while minimising the total distance.Operationally,the task is carried out by construc

167、ting a table,say X.Each element in the table,X,acrepresents the distance between household a from AIDIS and household c from CPHS datasets.The distance measure used for the Hungarian method is identical to that of the KNN.The lower the distance between the two households,the more similar they are.Th

168、e Hungarian method then iteratively selects pairs of unique households in a manner such that the sum of all distances(between two matched households)is minimised.We can consider an example to understand this better.Say there are two households,a and a from AIDIS and c,c,and c 12123from CPHS.Thus,the

169、re are six possible assignments:c a,c a,c a,c a,c a,c a.112131122232Firstly,the Hungarian method considers the assignment,c a,as a given(say,with a distance 11of 0.2).At this stage,both c and a are considered 11assigned,and thus,the model only computes the distance for c a (say,a distance of 0.3)and

170、 c?a 2232(say,a distance of 0.4),i.e.,the residual pairs.Thus,in the first iteration,the optimal match is found to be c?1a and c a,with a total distance of 0.5.The model 122then considers the pair c a as fixed and computes 21the distance for the residual pairs,which,let us say,results in a minimum t

171、otal distance of 0.4,with c 2a and c a representing the matches.Finally,in 112the third iteration of the model,c a will be 31considered fixed,and the distance of the residual pairs will be computed.Out of these three iterations,let us say the second iteration resulted in the lowest sum of distances.

172、In such a scenario,the resultant pair from the second iteration is considered final.Thus,combining the KNN and the Hungarian methods provides a comprehensive and effective approach to household matching.The formers 8 It can be intuitively understood in the following example:Assume we compare two cou

173、ntries based on one parameter,say GDP.Then,we are likely to find a difference.As we start adding dimensions,say population,growth rates,gender distribution,life expectancy,majority religion,etc.,in some cases,the distances will start increasing(e.g.,if we were comparing India and Bangladesh),while i

174、n others the distances will start reducing(e.g.,if we were comparing Iran and Turkey,which have similar population,life expectancy,and so on).So,as the number of variables(dimensions)increase,the chances that two countries may appear similar increases,especially when we are adding the difference in

175、the variables.The KNN method also has a few drawbacks(Guo et al.,2003).Firstly,its computational complexity increases with the size of the dataset(Maillo et al.,2015;Maillo et al.,2017;Deng et al.,2016).Secondly,in high-dimensional spaces where instances tend to 8be equidistant,a challenge arises,im

176、pacting the methods performance,known as the curse of dimensionality.Finally,the KNN method is sensitive to imbalanced datasets,potentially leading to biased predictions(Goyal,2022).In this estimation exercise,the first two drawbacks,computational complexity and distances in higher dimensional space

177、s,are mitigated by reducing the total observations and dimensions.Observation reduction was done by selecting one state at a time from both datasets,and dimension reduction was done by selecting only 9 common variables across both AIDIS and CPHS datasets.The third challenge that arises due to imbala

178、nced datasets,resulting in higher and lower density regions,remains.For example,we expect to find more households earning between 10,000 and 1,00,000 than between?10,00,000 and?10,90,000,despite the interval being equal.Thus,when all variables are considered together,regions of overpopulation(and hi

179、gher densities)and regions of underpopulation(and lower densities)emerge.This prevents us from achieving a 1:1(unique)match.To mitigate this hurdle,we also use the Hungarian Algorithm to find matching households between the two datasets.The Hungarian method,developed by Hungarian mathematicians Dnes

180、 Knig and Jen Egervry in the 1930s,has found applications in various fields.It solves the classification problem where the goal is to find the optimal assignment of a set of tasks to a set of agents,minimising the total cost(Hahn et al.,1998).In our context,the goal is to assign 0708flexibility in h

181、andling mixed variable types and adaptability to complex distributions,combined with the latters precision in achieving an optimal one-to-one mapping,creates a synergistic effect that addresses the discussed challenges in the matching process.To generate unique one-to-one mapping,we must match from

182、the smaller dataset to the bigger one meaning that for the states where AIDIS has the smaller number of households,we will try to find for each AIDIS household a corresponding and unique household from the CMIE dataset that is its closest match.Thus,to combine both models,we start with KNN.Assuming

183、that AIDIS has fewer households for all states compared to the CPHS,the KNN model shall result in some one-to-one matching(one household from the CPHS dataset will be assigned to one from AIDIS),some one-to-many matching(one household from CPHS will be assigned to many households of AIDIS),as well a

184、s some residual households(of CPHS who were not assigned to any households in AIDIS).These unique(one-to-one)matches are considered final matches.For the one-to-many matches,we consider the closest match as the final match.To exemplify,say,household c1 of CPHS was matched with households a,a,and a o

185、f AIDIS.The distance 123between each pair c-a,c-a,and c-a are 0.2,0.25 111213and 0.35,respectively.So,despite three matches,we only consider the c1-a1 pair since this has the lowest distance.We obtain a set of matched and unmatched households using these one-to-one matches and by resolving the one-t

186、o-many matches.These matched households are used for final analysis,whereas the unmatched households are then passed onto the Hungarian method for final matching.Section 3.2:Data TransformationsData cleaning is a crucial step in the pre-processing pipeline,especially when dealing with datasets that

187、include both categorical and continuous variables.Following are the key strategies adopted for data cleaning before employing the K-Nearest Neighbors(KNN)and the Hungarian method.l Handling Missing Values:KNN and the Hungarian methods are sensitive to missing data.Given the negligible occurrence of

188、such missing data across the variables used for matching and estimating the impact of the FSP,imputation methods are avoided since they may introduce bias or distort the original distribution.Instead,such households were dropped.l Standardising and Scaling:KNN relies on distance metrics,and the Hung

189、arian method involves optimisation,both of which are influenced by the scale of variables.Thus,observations were standardised by subtracting the minimum value and dividing by the range (maximum observed value minimum observed value of the variable).l Recasting Categorical Variables:Categorical varia

190、bles,wherever in the form of non-numeric values,were converted into a numerical format.l Ensuring Compatibility with Methods:Finally,since the two methods have specific requirements regarding the input data format,the datasets were reorganised and variables were appropriately pre-processed to ensure

191、 compatibility.Upon completion of the data transformation,the KNN and the Hungarian methods were used to obtain the final data structure based on which estimations were carried out.Before discussing the final data structure,it is important to discuss one final aspect of the matching procedure:the qu

192、antum of data loss.It is evident that whether the matching happens from AIDIS to CPHS or from CPHS to AIDIS,the final results will not differ since the final result will indicate that households a and c(from AIDIS and CPHS,respectively)are identical.However,the number of households in each state may

193、 differ.For example,in Bihar,AIDIS has 7708 households and CPHS has 9236 households,households from the CPHS dataset to households in the AIDIS dataset while minimising the total distance.Operationally,the task is carried out by constructing a table,say X.Each element in the table,X,acrepresents the

194、 distance between household a from AIDIS and household c from CPHS datasets.The distance measure used for the Hungarian method is identical to that of the KNN.The lower the distance between the two households,the more similar they are.The Hungarian method then iteratively selects pairs of unique hou

195、seholds in a manner such that the sum of all distances(between two matched households)is minimised.We can consider an example to understand this better.Say there are two households,a and a from AIDIS and c,c,and c 12123from CPHS.Thus,there are six possible assignments:c a,c a,c a,c a,c a,c a.1121311

196、22232Firstly,the Hungarian method considers the assignment,c a,as a given(say,with a distance 11of 0.2).At this stage,both c and a are considered 11assigned,and thus,the model only computes the distance for c a (say,a distance of 0.3)and c?a 2232(say,a distance of 0.4),i.e.,the residual pairs.Thus,i

197、n the first iteration,the optimal match is found to be c?1a and c a,with a total distance of 0.5.The model 122then considers the pair c a as fixed and computes 21the distance for the residual pairs,which,let us say,results in a minimum total distance of 0.4,with c 2a and c a representing the matches

198、.Finally,in 112the third iteration of the model,c a will be 31considered fixed,and the distance of the residual pairs will be computed.Out of these three iterations,let us say the second iteration resulted in the lowest sum of distances.In such a scenario,the resultant pair from the second iteration

199、 is considered final.Thus,combining the KNN and the Hungarian methods provides a comprehensive and effective approach to household matching.The formers 8 It can be intuitively understood in the following example:Assume we compare two countries based on one parameter,say GDP.Then,we are likely to fin

200、d a difference.As we start adding dimensions,say population,growth rates,gender distribution,life expectancy,majority religion,etc.,in some cases,the distances will start increasing(e.g.,if we were comparing India and Bangladesh),while in others the distances will start reducing(e.g.,if we were comp

201、aring Iran and Turkey,which have similar population,life expectancy,and so on).So,as the number of variables(dimensions)increase,the chances that two countries may appear similar increases,especially when we are adding the difference in the variables.The KNN method also has a few drawbacks(Guo et al

202、.,2003).Firstly,its computational complexity increases with the size of the dataset(Maillo et al.,2015;Maillo et al.,2017;Deng et al.,2016).Secondly,in high-dimensional spaces where instances tend to 8be equidistant,a challenge arises,impacting the methods performance,known as the curse of dimension

203、ality.Finally,the KNN method is sensitive to imbalanced datasets,potentially leading to biased predictions(Goyal,2022).In this estimation exercise,the first two drawbacks,computational complexity and distances in higher dimensional spaces,are mitigated by reducing the total observations and dimensio

204、ns.Observation reduction was done by selecting one state at a time from both datasets,and dimension reduction was done by selecting only 9 common variables across both AIDIS and CPHS datasets.The third challenge that arises due to imbalanced datasets,resulting in higher and lower density regions,rem

205、ains.For example,we expect to find more households earning between 10,000 and 1,00,000 than between?10,00,000 and?10,90,000,despite the interval being equal.Thus,when all variables are considered together,regions of overpopulation(and higher densities)and regions of underpopulation(and lower densiti

206、es)emerge.This prevents us from achieving a 1:1(unique)match.To mitigate this hurdle,we also use the Hungarian Algorithm to find matching households between the two datasets.The Hungarian method,developed by Hungarian mathematicians Dnes Knig and Jen Egervry in the 1930s,has found applications in va

207、rious fields.It solves the classification problem where the goal is to find the optimal assignment of a set of tasks to a set of agents,minimising the total cost(Hahn et al.,1998).In our context,the goal is to assign 0708flexibility in handling mixed variable types and adaptability to complex distri

208、butions,combined with the latters precision in achieving an optimal one-to-one mapping,creates a synergistic effect that addresses the discussed challenges in the matching process.To generate unique one-to-one mapping,we must match from the smaller dataset to the bigger one meaning that for the stat

209、es where AIDIS has the smaller number of households,we will try to find for each AIDIS household a corresponding and unique household from the CMIE dataset that is its closest match.Thus,to combine both models,we start with KNN.Assuming that AIDIS has fewer households for all states compared to the

210、CPHS,the KNN model shall result in some one-to-one matching(one household from the CPHS dataset will be assigned to one from AIDIS),some one-to-many matching(one household from CPHS will be assigned to many households of AIDIS),as well as some residual households(of CPHS who were not assigned to any

211、 households in AIDIS).These unique(one-to-one)matches are considered final matches.For the one-to-many matches,we consider the closest match as the final match.To exemplify,say,household c1 of CPHS was matched with households a,a,and a of AIDIS.The distance 123between each pair c-a,c-a,and c-a are 0

212、.2,0.25 111213and 0.35,respectively.So,despite three matches,we only consider the c1-a1 pair since this has the lowest distance.We obtain a set of matched and unmatched households using these one-to-one matches and by resolving the one-to-many matches.These matched households are used for final anal

213、ysis,whereas the unmatched households are then passed onto the Hungarian method for final matching.Section 3.2:Data TransformationsData cleaning is a crucial step in the pre-processing pipeline,especially when dealing with datasets that include both categorical and continuous variables.Following are

214、 the key strategies adopted for data cleaning before employing the K-Nearest Neighbors(KNN)and the Hungarian method.l Handling Missing Values:KNN and the Hungarian methods are sensitive to missing data.Given the negligible occurrence of such missing data across the variables used for matching and es

215、timating the impact of the FSP,imputation methods are avoided since they may introduce bias or distort the original distribution.Instead,such households were dropped.l Standardising and Scaling:KNN relies on distance metrics,and the Hungarian method involves optimisation,both of which are influenced

216、 by the scale of variables.Thus,observations were standardised by subtracting the minimum value and dividing by the range (maximum observed value minimum observed value of the variable).l Recasting Categorical Variables:Categorical variables,wherever in the form of non-numeric values,were converted

217、into a numerical format.l Ensuring Compatibility with Methods:Finally,since the two methods have specific requirements regarding the input data format,the datasets were reorganised and variables were appropriately pre-processed to ensure compatibility.Upon completion of the data transformation,the K

218、NN and the Hungarian methods were used to obtain the final data structure based on which estimations were carried out.Before discussing the final data structure,it is important to discuss one final aspect of the matching procedure:the quantum of data loss.It is evident that whether the matching happ

219、ens from AIDIS to CPHS or from CPHS to AIDIS,the final results will not differ since the final result will indicate that households a and c(from AIDIS and CPHS,respectively)are identical.However,the number of households in each state may differ.For example,in Bihar,AIDIS has 7708 households and CPHS

220、 has 9236 households,Section 3.3:Final Data Structure The final dataset contains all the variables used for merging,along with additional variables from the AIDIS and CMIE datasets.Table-3 presents the 10description of the variables and their source data:Estimations were done using these variables f

221、or the households across Indian states and union territories.The analysis however excludes Andaman&Nicobar Islands,Arunachal Pradesh,Dadra&Nagar Haveli,Daman&Diu,Lakshadweep,Manipur,Mizoram,and Nagaland as the CMIE CPHS does not report data for those states in 2019.In addition to sample-level estima

222、tions,we also use the weights provided by the two datasets to project the estimations onto the population level.For states where the base dataset is AIDIS,i.e.,where all households of AIDIS are assigned a corresponding household from the CPHS dataset,we use the weights in the AIDIS dataset to comput

223、e state-population-level results.Similarly,for states where the CPHS dataset is used as a base dataset,CPHS weights are used.For most states,we rely on the AIDIS dataset as the base dataset due to its smaller state-specific sample size.9 9236(households in CPHS)-7708(households in AIDIS)=1528 Househ

224、olds from CPHS who were not assigned a corresponding household from the AIDIS dataset.10 The total number of variables used for the estimation is 156,but between them they contain the data pertaining to the themes discussed in the table.All 156 variables are not reproduced here to enhance ease of un

225、derstanding.9and thus,1528 households from CPHS do not get any households from AIDIS assigned to them.The data pertaining to these(1528 in case of Bihar)residual households are thus not accounted for in the final dataset.Appendix-A presents the number of households that were residual households for

226、each of the analysed states.Table-3:Variables present in the final data(used for estimations)Source:Authors Calculations From CMIE YesYesYesYesYesYesYesYesYes-YesYesFrom AIDIS YesYesYesYesYesYesYesYesYesYesYes-Sl 12345678910111213Variable Name Region(Urban/Rural)District Social Group Religion Age Gr

227、oups Gender Groups Household(HH)Size HH Expenditure#Similar HHs in the State Value of assets owned by the HH(across various types of assets)Amount of Debt Outstanding Occupational Sector of the Head of the HH Household Income 0910In case of AIDIS,weights are assigned at the stratum or district level

228、.To compute the total number of FSP-eligible households in the population,we identify qualifying households in the sample,multiply their eligibility by the assigned weight,and sum up these values for a population-level estimate(National Sample Survey Organisation,2019).However,for Assam,Delhi,Meghal

229、aya,Sikkim,and Tripura,we turn to the CMIE CPHS as the base dataset.When using CMIE CPHS as the base,we apply the datasets provided weights,utilising state-level weights for households and a non-response factor.The weight of an observation is calculated by scaling the state-level weight with the non

230、-response factor,yielding a measure for each household per month.These constructed weights are averaged to derive a final measure for each household in the year 2019,which is then employed for all population-level estimates(Consumer Pyramids Household Survey,2019).The estimation results are discusse

231、d in the next section.Section 3.3:Final Data Structure The final dataset contains all the variables used for merging,along with additional variables from the AIDIS and CMIE datasets.Table-3 presents the 10description of the variables and their source data:Estimations were done using these variables

232、for the households across Indian states and union territories.The analysis however excludes Andaman&Nicobar Islands,Arunachal Pradesh,Dadra&Nagar Haveli,Daman&Diu,Lakshadweep,Manipur,Mizoram,and Nagaland as the CMIE CPHS does not report data for those states in 2019.In addition to sample-level estim

233、ations,we also use the weights provided by the two datasets to project the estimations onto the population level.For states where the base dataset is AIDIS,i.e.,where all households of AIDIS are assigned a corresponding household from the CPHS dataset,we use the weights in the AIDIS dataset to compu

234、te state-population-level results.Similarly,for states where the CPHS dataset is used as a base dataset,CPHS weights are used.For most states,we rely on the AIDIS dataset as the base dataset due to its smaller state-specific sample size.9 9236(households in CPHS)-7708(households in AIDIS)=1528 House

235、holds from CPHS who were not assigned a corresponding household from the AIDIS dataset.10 The total number of variables used for the estimation is 156,but between them they contain the data pertaining to the themes discussed in the table.All 156 variables are not reproduced here to enhance ease of u

236、nderstanding.9and thus,1528 households from CPHS do not get any households from AIDIS assigned to them.The data pertaining to these(1528 in case of Bihar)residual households are thus not accounted for in the final dataset.Appendix-A presents the number of households that were residual households for

237、 each of the analysed states.Table-3:Variables present in the final data(used for estimations)Source:Authors Calculations From CMIE YesYesYesYesYesYesYesYesYes-YesYesFrom AIDIS YesYesYesYesYesYesYesYesYesYesYes-Sl 12345678910111213Variable Name Region(Urban/Rural)District Social Group Religion Age G

238、roups Gender Groups Household(HH)Size HH Expenditure#Similar HHs in the State Value of assets owned by the HH(across various types of assets)Amount of Debt Outstanding Occupational Sector of the Head of the HH Household Income 0910In case of AIDIS,weights are assigned at the stratum or district leve

239、l.To compute the total number of FSP-eligible households in the population,we identify qualifying households in the sample,multiply their eligibility by the assigned weight,and sum up these values for a population-level estimate(National Sample Survey Organisation,2019).However,for Assam,Delhi,Megha

240、laya,Sikkim,and Tripura,we turn to the CMIE CPHS as the base dataset.When using CMIE CPHS as the base,we apply the datasets provided weights,utilising state-level weights for households and a non-response factor.The weight of an observation is calculated by scaling the state-level weight with the no

241、n-response factor,yielding a measure for each household per month.These constructed weights are averaged to derive a final measure for each household in the year 2019,which is then employed for all population-level estimates(Consumer Pyramids Household Survey,2019).The estimation results are discuss

242、ed in the next section.IV.Estimation Results Households were matched using a tiered approach.The first layer of matching was done using KNN,and the second layer using the Hungarian model.The following table,Table-4,presents the total number of households(of AIDIS)matched in each stage and their mean

243、 distances.Figures-1(A)and-1(B)present the distribution of distances of the matched households across the two methods.Figure-1:Distribution of distances between matched households using KNN(A)and Hungarian Method(B)Table-4:Households matched through each model(and summary statistics of the distances

244、)Median Distance(Modified Gower)Mean Distance(Modified Gower)0.080.090.160.18Std.Dev.#HH(fromAIDIS/CMIE)Matched 0.05537323 0.07169093 KNN Method Hungarian Method Source:Authors Calculations Source:Authors Calculations As discussed in the earlier section,any pair of matched households will have two d

245、istancesone combined distance for categorical variables and one combined distance for continuous variables.Given that we summed the distance of all categorical variables and then divided it by 7,we obtained a stepwise distribution for categorical variables(between 0 and 1).Similarly,we obtain a cont

246、inuous distribution(between 0 and 1)for continuous While the summary statistics presented above are for the sample,the estimation results have been calculated for the population level by applying appropriate weights,as described in the previous Table-5:Summary statistics of the relevant variables fo

247、r determining eligibility under the FSP Source:Authors Calculations Count 1,06,416 50,058 1,06,416 1,06,416 Mean 2,41,405.7 3,19,361.3 23,67,650 0.83 StandardDeviation 2,09,040.5 10,36,296 85,67,444NAst1 quartile(Q)1,21,081.534,5702,72,175NAnd2 Q(Median)1,81,735 87,0008,95,000NArd3 Q 2,92,447 2,67,6

248、5923,76,150 NAVariableTotal Annual Income Outstanding Debt Value of Assets 11 Home OwnershipTable-6:Number of households qualifying for FSP under each of the eligibility criteria Source:Authors Calculations Qualifying Households12(from Matched Dataset)45,02,187 2,17,58,764 40,24,937 92,89,643 1,50,4

249、08 FSP Criterion-1:Annual Income?60,000 FSP Criterion-2:Outstanding debt amount?0 FSP Criterion-3:Value of Assets?20,000 FSP Criterion-4:No home ownership Combining all criteria 11 Home ownership is a categorical value.The mean is represented since it presents the ratio of number of people who own a

250、 residential property(from the data it appears that 94%of the sample owns a residential property).12 The following results have been calculated only for households that have reported owing some debt.variables.Thus,Figures 1(A)and 1(B)suggest that most of the matched households were fairly close to o

251、ne another.Using the merged data,we estimate the eligibility of the households for FSP.Table-5 presents the summary statistics of the relevant variables(at the sample level).section.Table-6 presents how many households qualify under each of the four criteria laid out for the FSP.1112IV.Estimation Re

252、sults Households were matched using a tiered approach.The first layer of matching was done using KNN,and the second layer using the Hungarian model.The following table,Table-4,presents the total number of households(of AIDIS)matched in each stage and their mean distances.Figures-1(A)and-1(B)present

253、the distribution of distances of the matched households across the two methods.Figure-1:Distribution of distances between matched households using KNN(A)and Hungarian Method(B)Table-4:Households matched through each model(and summary statistics of the distances)Median Distance(Modified Gower)Mean Di

254、stance(Modified Gower)0.080.090.160.18Std.Dev.#HH(fromAIDIS/CMIE)Matched 0.05537323 0.07169093 KNN Method Hungarian Method Source:Authors Calculations Source:Authors Calculations As discussed in the earlier section,any pair of matched households will have two distancesone combined distance for categ

255、orical variables and one combined distance for continuous variables.Given that we summed the distance of all categorical variables and then divided it by 7,we obtained a stepwise distribution for categorical variables(between 0 and 1).Similarly,we obtain a continuous distribution(between 0 and 1)for

256、 continuous While the summary statistics presented above are for the sample,the estimation results have been calculated for the population level by applying appropriate weights,as described in the previous Table-5:Summary statistics of the relevant variables for determining eligibility under the FSP

257、 Source:Authors Calculations Count 1,06,416 50,058 1,06,416 1,06,416 Mean 2,41,405.7 3,19,361.3 23,67,650 0.83 StandardDeviation 2,09,040.5 10,36,296 85,67,444NAst1 quartile(Q)1,21,081.534,5702,72,175NAnd2 Q(Median)1,81,735 87,0008,95,000NArd3 Q 2,92,447 2,67,65923,76,150 NAVariableTotal Annual Inco

258、me Outstanding Debt Value of Assets 11 Home OwnershipTable-6:Number of households qualifying for FSP under each of the eligibility criteria Source:Authors Calculations Qualifying Households12(from Matched Dataset)45,02,187 2,17,58,764 40,24,937 92,89,643 1,50,408 FSP Criterion-1:Annual Income?60,000

259、 FSP Criterion-2:Outstanding debt amount?0 FSP Criterion-3:Value of Assets?20,000 FSP Criterion-4:No home ownership Combining all criteria 11 Home ownership is a categorical value.The mean is represented since it presents the ratio of number of people who own a residential property(from the data it

260、appears that 94%of the sample owns a residential property).12 The following results have been calculated only for households that have reported owing some debt.variables.Thus,Figures 1(A)and 1(B)suggest that most of the matched households were fairly close to one another.Using the merged data,we est

261、imate the eligibility of the households for FSP.Table-5 presents the summary statistics of the relevant variables(at the sample level).section.Table-6 presents how many households qualify under each of the four criteria laid out for the FSP.1112Combining all four criteria,we find that only 1,50,408

262、out of the 26,56,71,317 households with outstanding debt qualify for FSP.This represents 0.057%of all households.The number of qualifying households Table-7:Share of households qualifying for FSP under the income criteria Base weights used Total CMIE CPHS22525.75 AIDISCMIE CPHSCMIE CPHSCMIE CPHSAIDI

263、SCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSAIDISCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSAIDISCMIE CPHSCMIE CPHSAIDISCMIE CPHSCMIE CPHSCMIE CPHS1115 554.83 000012.5 000198.5 304.5 2663.25 6806 3692.5 546 63726.63 0515.25 02175 27230.17 0010009.25 08332.58 1

264、50407.71%State Count Qualifying HHs(Expenditure criterion)0.1586 14198806 0.0091 0.0031 0.0000 0.0000 0.0000 0.0000 0.00010.0000 0.00000.0000 0.0030 0.0022 0.0299 0.0455 0.0152 0.0709 0.6363 0.0000 0.0086 0.00000.3915 0.1421 0.0000 0.0000 0.0285 0.0000 0.0373 0.0566 12200471 17748050 252275.0938 567

265、2758.5 4922844 308249.5 12531386 5414255.5 1716132.75 2272021.25 6516384.5 13810240 8910524 14949053 24223068 770592 10015405 288658 6019335 13273183 555504 19161852 9276134 1261376 35141980 1947767 22313012 265671317.1 Andhra Pradesh AssamBihar Chandigarh Chhattisgarh Delhi GoaGujarat Haryana Himac

266、hal Pradesh Jammu&Kashmir Jharkhand Karnataka Kerala Madhya Pradesh Maharashtra Meghalaya Odisha Puducherry Punjab Rajasthan Sikkim Tamil Nadu Telangana Tripura Uttar Pradesh Uttarakhand West Bengal Total Source:Authors Calculations across each state is represented in Appendix C.Table-7 presents the

267、 number and proportions of qualifying households at the state level.We also explore an alternative estimation approach.Earlier,we used four criteria(given in Table-8).However,if we replace criterion-1,i.e.,the income of the household must be less than 60,000 annually,with expenditure of the househol

268、d must be less than 60,000 annually,we find that the number of households that qualify for FSP increases from 1,50,408 to 4,42,802.We construct this scenario(by replacing income with expenditure)since most measures of poverty focus on the expenditure of the individual or household rather than income

269、.Table-8 provides the number of households that qualify for this revised criteria.The estimates reveal that Odisha(with 63,727 households),Tamil Nadu(27,230 households),and Andhra Pradesh(22,526 households)are the states with the highest number of households that qualify for FSP.Together,these state

270、s account for 75%of the total number of qualifying households per the income criterion.These states also constitute 77%of the total outstanding debt that qualifies for FSP.Further,there are twelve states without any qualifying households.The estimation results thus suggest that there are pockets of

271、concentration where FSP may have a higher uptake,assuming the ratio of qualifying households vis-vis households that seek refuge remains constant across regions,states,and cultures.Combining the revised criteria(replacing income with expenditure),we find that only 4,42,802 households out of the 26,5

272、6,71,317 households with outstanding debt qualify for FSP,i.e.,only 0.166%of households qualify for FSP.Table-9 presents the state-level qualifications.Under the revised criteria,Odisha still has 1,03,537 qualifying households,which is the highest in the country.It is followed by West Bengal with 96

273、,159 and Uttar Pradesh with 54,641 qualifying households.These three states together account for 57%of the Table-8:Number of households qualifying for FSP under the revised criteria(expenditure-based)Source:Authors Calculations Qualifying Households13(from Matched Dataset)1,04,05,050 2,17,58,764 40,

274、24,937 92,89,643 4,42,802 FSP Criterion-1:Annual Income?60,000 FSP Criterion-2:Outstanding debt amount?0 FSP Criterion-3:Value of Assets?20,000 FSP Criterion-4:No home ownership Combining all criteria total number of qualifying households and 56%of the total qualifying outstanding debt,considering t

275、he expenditure criterion(alongside asset,debt and home ownership criteria).In this scenario,the number of states with zero qualifying households comes down to six.The number of qualifying households across each state is represented in Appendix D.Table-9 presents the number and proportions of qualify

276、ing households at the state level.131413 The following results have been calculated only for households that have reported owing some debt.Combining all four criteria,we find that only 1,50,408 out of the 26,56,71,317 households with outstanding debt qualify for FSP.This represents 0.057%of all hous

277、eholds.The number of qualifying households Table-7:Share of households qualifying for FSP under the income criteria Base weights used Total CMIE CPHS22525.75 AIDISCMIE CPHSCMIE CPHSCMIE CPHSAIDISCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSCMIE CPHSAIDISCMIE CPHSC

278、MIE CPHSCMIE CPHSCMIE CPHSAIDISCMIE CPHSCMIE CPHSAIDISCMIE CPHSCMIE CPHSCMIE CPHS1115 554.83 000012.5 000198.5 304.5 2663.25 6806 3692.5 546 63726.63 0515.25 02175 27230.17 0010009.25 08332.58 150407.71%State Count Qualifying HHs(Expenditure criterion)0.1586 14198806 0.0091 0.0031 0.0000 0.0000 0.00

279、00 0.0000 0.00010.0000 0.00000.0000 0.0030 0.0022 0.0299 0.0455 0.0152 0.0709 0.6363 0.0000 0.0086 0.00000.3915 0.1421 0.0000 0.0000 0.0285 0.0000 0.0373 0.0566 12200471 17748050 252275.0938 5672758.5 4922844 308249.5 12531386 5414255.5 1716132.75 2272021.25 6516384.5 13810240 8910524 14949053 24223

280、068 770592 10015405 288658 6019335 13273183 555504 19161852 9276134 1261376 35141980 1947767 22313012 265671317.1 Andhra Pradesh AssamBihar Chandigarh Chhattisgarh Delhi GoaGujarat Haryana Himachal Pradesh Jammu&Kashmir Jharkhand Karnataka Kerala Madhya Pradesh Maharashtra Meghalaya Odisha Puducherr

281、y Punjab Rajasthan Sikkim Tamil Nadu Telangana Tripura Uttar Pradesh Uttarakhand West Bengal Total Source:Authors Calculations across each state is represented in Appendix C.Table-7 presents the number and proportions of qualifying households at the state level.We also explore an alternative estimat

282、ion approach.Earlier,we used four criteria(given in Table-8).However,if we replace criterion-1,i.e.,the income of the household must be less than 60,000 annually,with expenditure of the household must be less than 60,000 annually,we find that the number of households that qualify for FSP increases f

283、rom 1,50,408 to 4,42,802.We construct this scenario(by replacing income with expenditure)since most measures of poverty focus on the expenditure of the individual or household rather than income.Table-8 provides the number of households that qualify for this revised criteria.The estimates reveal tha

284、t Odisha(with 63,727 households),Tamil Nadu(27,230 households),and Andhra Pradesh(22,526 households)are the states with the highest number of households that qualify for FSP.Together,these states account for 75%of the total number of qualifying households per the income criterion.These states also c

285、onstitute 77%of the total outstanding debt that qualifies for FSP.Further,there are twelve states without any qualifying households.The estimation results thus suggest that there are pockets of concentration where FSP may have a higher uptake,assuming the ratio of qualifying households vis-vis house

286、holds that seek refuge remains constant across regions,states,and cultures.Combining the revised criteria(replacing income with expenditure),we find that only 4,42,802 households out of the 26,56,71,317 households with outstanding debt qualify for FSP,i.e.,only 0.166%of households qualify for FSP.Ta

287、ble-9 presents the state-level qualifications.Under the revised criteria,Odisha still has 1,03,537 qualifying households,which is the highest in the country.It is followed by West Bengal with 96,159 and Uttar Pradesh with 54,641 qualifying households.These three states together account for 57%of the

288、 Table-8:Number of households qualifying for FSP under the revised criteria(expenditure-based)Source:Authors Calculations Qualifying Households13(from Matched Dataset)1,04,05,050 2,17,58,764 40,24,937 92,89,643 4,42,802 FSP Criterion-1:Annual Income?60,000 FSP Criterion-2:Outstanding debt amount?0 F

289、SP Criterion-3:Value of Assets 02055244211512178552.38092.5653087.142185863191.6160018.2130411.236834.56278.7172585.01620216.6581543.7279502658418.543800500311.82709.67162778.6213954.1323261272215662977.919518539266.463116.09478225.4Debt=0300570.2324608191464.40188878.20171.539374.675066.221286.8312

290、12.5158298.4142331.147264.92213969.3256065.554663742932429740.5126264.211707179500.7119193.912121279559.42849.17691942Debt0933535.9378199628172.3418249048.10137101095.817507.828279.8532003.25252192.7615740.6111829.31095408878256.1304813728012672.889158.84201695.710311549265.9716122.82744314111411713

291、6.5762429.4Debt=0123232.1754423130923.6054171.540020334.68002713.548890.9225612.336633.5812866.9393447.331092265615.103225.7534057.815813687498.6916969.08546072191.382118.67314548.3Debt0418175.3708745303135.4067487.820978.554626.3548994141685.1549006.3210727811789495783.74331163.82472740722.8195.547

292、26.4689070.67101262272428.7189609.93798427222.53870.5369419State Andhra Pradesh AssamBihar Chandigarh Chhattisgarh Delhi GoaGujarat Haryana Himachal Pradesh Jammu&Kashmir Jharkhand Karnataka Kerala Madhya Pradesh Maharashtra Meghalaya Odisha Puducherry Punjab Rajasthan Sikkim Tamil Nadu Telangana Tr

293、ipura Uttar Pradesh Uttarakhand West Bengal Total 265671317.14211739928964321641624502187396173910405049.72Source:Authors Calculations Debt0DebtAsset ValueTotal%OutstandingDebtTotalQualifying HHs(Expenditure criteria)Qualifying HHs(Income criteria)%OutstandingDebt10385.40112107822525.750.15865637073

294、34443708.010.30782963700992067845163080011150.0091427643080445660.36528171658592051908.92250797554.830.00313815278011562.830.06515109707776447.334302.1800.0000004180.165692627000032097.5475626.500.000000690.0012167590009164412147300.000000000960364200.0000001370.044445274000036691.82516420.112.50.00

295、0101250001754.50.0140012025095034045.22242019.900.0000002265.50.041843217263005089.2362388.1100.000000000934.7596731.6400.00000000018491.62723043.1198.50.003054597002658.50.04079759063076203080.7873674.8304.50.0022030937508314.060.060202139030528191504.9528871.12663.250.02989496626761013.250.0113712

296、980367697341.1129814868060.045531622865121816.140.01214941296864222628.512519013692.50.015245761500017516.920.072315443746880172441080905460.070857985250000257125.5190021563726.630.63629992387584103536.91.03377616883979527060.2920554.1300.00000000083828.5415468.9515.250.00856891434059460.09878272510

297、00092092.61954387.300.0000004260.00320910692000155185050821750.391544716675212620.22718124341750503439.9124383427230.170.1421139880694434779.090.181502530335648242215.1746335.800.0000007051.250.076015190397104389915402100.000000000313916.2263248410009.250.0284815565884854641.190.15548753876608014550

298、.58110704.200.00000032010.16434267219416Debt=03036458060822422.551179.111377.51523.55921.3353752.3528060.266870.547531440.5251237.6582075.4339889.98145866.56324113689.6025173.7142575.7420513272762.769181.719224114561.13716.33335395.3380796.822212458332.580.037348745297696159.30.430956170483878420252

299、94402493721758764150407.70.056612644484536442802.40.166673705548962422Note:The figures presented in the table above are for the population-level.It lays out the number of households that qualify for FSP for every criterion,namely,home ownership,total income,asset value,and outstanding debt.We also d

300、o a similar calculation for total expenditure.We then calculate the final number of households that would qualify if all the criteria were to be applied.Although the calculations under outstanding debt and final income and expenditure criteria only account for households that owe some debt,for the o

301、ther calculations we present the figures for households that currently do not owe any debt as there is a chance that they may become indebted in the future21Appendix B:Detailed table(non-rounded)on the share of qualifying householdsunder FSP(income and expenditure criteria)Home OwnershipTotal Expend

302、itureTotal IncomeCount141988061220047117748050252275.09385672758.54922844308249.5125313865414255.51716132.752272021.256516384.5138102408910524149490532422306877059210015405288658601933513273183555504191618529276134126137635141980194776722313012Debt=0635071.121853362322.272957.3538101.922792972481.16

303、106929.152494.379061.751644.3346793.54130869.6193367.796381.99284476.48505217898.57653.2167820.64159006.726312675535.118649015349189551.913561.16483270.4Debt02055244211512178552.38092.5653087.142185863191.6160018.2130411.236834.56278.7172585.01620216.6581543.7279502658418.543800500311.82709.67162778

304、.6213954.1323261272215662977.919518539266.463116.09478225.4Debt=0300570.2324608191464.40188878.20171.539374.675066.221286.831212.5158298.4142331.147264.92213969.3256065.554663742932429740.5126264.211707179500.7119193.912121279559.42849.17691942Debt0933535.9378199628172.3418249048.10137101095.817507.

305、828279.8532003.25252192.7615740.6111829.31095408878256.1304813728012672.889158.84201695.710311549265.9716122.827443141114117136.5762429.4Debt=0123232.1754423130923.6054171.540020334.68002713.548890.9225612.336633.5812866.9393447.331092265615.103225.7534057.815813687498.6916969.08546072191.382118.673

306、14548.3Debt0418175.3708745303135.4067487.820978.554626.3548994141685.1549006.3210727811789495783.74331163.82472740722.8195.54726.4689070.67101262272428.7189609.93798427222.53870.5369419State Andhra Pradesh AssamBihar Chandigarh Chhattisgarh Delhi GoaGujarat Haryana Himachal Pradesh Jammu&Kashmir Jha

307、rkhand Karnataka Kerala Madhya Pradesh Maharashtra Meghalaya Odisha Puducherry Punjab Rajasthan Sikkim Tamil Nadu Telangana Tripura Uttar Pradesh Uttarakhand West Bengal Total 265671317.14211739928964321641624502187396173910405049.72Source:Authors Calculations Debt0DebtAsset ValueTotal%OutstandingDe

308、btTotalQualifying HHs(Expenditure criteria)Qualifying HHs(Income criteria)%OutstandingDebt10385.40112107822525.750.1586563707334443708.010.30782963700992067845163080011150.0091427643080445660.36528171658592051908.92250797554.830.00313815278011562.830.06515109707776447.334302.1800.0000004180.16569262

309、7000032097.5475626.500.000000690.0012167590009164412147300.000000000960364200.0000001370.044445274000036691.82516420.112.50.000101250001754.50.0140012025095034045.22242019.900.0000002265.50.041843217263005089.2362388.1100.000000000934.7596731.6400.00000000018491.62723043.1198.50.003054597002658.50.0

310、4079759063076203080.7873674.8304.50.0022030937508314.060.060202139030528191504.9528871.12663.250.02989496626761013.250.0113712980367697341.1129814868060.045531622865121816.140.01214941296864222628.512519013692.50.015245761500017516.920.072315443746880172441080905460.070857985250000257125.51900215637

311、26.630.63629992387584103536.91.03377616883979527060.2920554.1300.00000000083828.5415468.9515.250.00856891434059460.0987827251000092092.61954387.300.0000004260.00320910692000155185050821750.391544716675212620.22718124341750503439.9124383427230.170.1421139880694434779.090.181502530335648242215.1746335

312、.800.0000007051.250.076015190397104389915402100.000000000313916.2263248410009.250.0284815565884854641.190.15548753876608014550.58110704.200.00000032010.16434267219416Debt=03036458060822422.551179.111377.51523.55921.3353752.3528060.266870.547531440.5251237.6582075.4339889.98145866.56324113689.6025173

313、.7142575.7420513272762.769181.719224114561.13716.33335395.3380796.822212458332.580.037348745297696159.30.43095617048387842025294402493721758764150407.70.056612644484536442802.40.16667370554896242223Appendix C:State-wise number of qualifying HHs(Income criterion)Appendix D:State-wise number of qualif

314、ying HHs(Expenditure criterion)Source:Authors Calculations Source:Authors Calculations Appendix E:State-wise amount of qualifying debt in INR Crores(Income criteria)24Source:Authors Calculations Appendix F:State-wise amount of qualifying debt(Expenditure criteria)Source:Authors Calculations 23Append

315、ix C:State-wise number of qualifying HHs(Income criterion)Appendix D:State-wise number of qualifying HHs(Expenditure criterion)Source:Authors Calculations Source:Authors Calculations Appendix E:State-wise amount of qualifying debt in INR Crores(Income criteria)24Source:Authors Calculations Appendix

316、F:State-wise amount of qualifying debt(Expenditure criteria)Source:Authors Calculations 25References Abu Alfeilat,H.A.,Hassanat,A.B.,Lasassmeh,O.,Tarawneh,A.S.,Alhasanat,M.B.,Eyal Salman,H.S.,&Prasath,V.S.(2019).Effects of distance measure choice on k-nearest neighbor classifier performance:a review

317、.Big data,7(4),221-248.Bhattacharya,D.,&Ghosh,I.(2022).Must Shylock Receive his Pound of Flesh?An Ethical Inquiry into the Insolvency and Bankruptcy Regimes for Natural Persons.https:/ibbi.gov.in/uploads/whatsnew/b5fba368fbd5c5817333f95fbb0d48bb.pdf.Bhattacharya,D.,&Ananth,B.(2021).The Curious Case

318、of Fresh Start Process:An inquiry into historical antecedents,international analogues,and domestic substitutes.https:/ibbi.gov.in/uploads/whatsnew/1d8b31fc65f7ac6f09a973be8f12f868.pdfCarlos A.et al.(2019),“Bankruptcy,Discharge,and the Emergence of Debtor Rights in Eighteenth-Century England.”Enterpr

319、ise&Society,Vol.20(2)pp.457-506.Chomboon,K.,Chujai,P.,Teerarassamee,P.,Kerdprasop,K.,&Kerdprasop,N.(2015,March).An empirical study of distance metrics for k-nearest neighbor algorithm.In Proceedings of the 3rd international conference on industrial application engineering(Vol.2).Consumer Pyramids Dx

320、.(2019).Consumer Pyramids Household Survey:Weights.Retrieved from https:/Cover,T.,&Hart,P.(1967).Nearest neighbor pattern classification.IEEE transactions on information theory,13(1),21-27.Deng,Z.,Zhu,X.,Cheng,D.,Zong,M.,&Zhang,S.(2016).Efficient kNN classification algorithm for big data.Neurocomput

321、ing,195,143-148.Dezan Shira&Associates(2023),“A Guide to Minimum Wage in India in 2023”,India Briefing,26 October.DOrazio,M.(2021).Distances with mixed type variables some modified Gowers coefficients.arXiv preprint arXiv:2101.02481.Fix,E.,&Hodges,J.L.(1989).Discriminatory analysis.Nonparametric dis

322、crimination:Consistency properties.International Statistical Review/Revue Internationale de Statistique,57(3),238-247.Government of India.(2016,May 28).Insolvency and Bankruptcy Code.https:/ibbi.gov.in/webadmin/pdf/legalframwork/2017/Jul/IBC%202016.pdf.Gower,J.C.(1971).A general coefficient of simil

323、arity and some of its properties.Biometrics,857-871.Citation count:6086(Source:Google Scholar)Goyal,S.(2022).Handling class-imbalance with KNN(neighbourhood)under-sampling for software defect prediction.Artificial Intelligence Review,55(3),2023-2064.26Guo,G.,Wang,H.,Bell,D.,Bi,Y.,&Greer,K.(2003).KNN

324、 model-based approach in classification.In On The Move to Meaningful Internet Systems 2003:CoopIS,DOA,and ODBASE:OTM Confederated International Conferences,CoopIS,DOA,and ODBASE 2003,Catania,Sicily,Italy,November 3-7,2003.Proceedings(pp.986-996).Springer Berlin Heidelberg.Hahn,P.,Grant,T.,&Hall,N.(1

325、998).A branch-and-bound algorithm for the quadratic assignment problem based on the Hungarian method.European Journal of Operational Research,108(3),629-640.Hu,L.Y.,Huang,M.W.,Ke,S.W.,&Tsai,C.F.(2016).The distance function effect on k-nearest neighbor classification for medical datasets.SpringerPlus

326、,5(1),1-9.Levinthal L.(1918),“The Early History of Bankruptcy Law”University of Pennsylvania Law Review and American Law Register,Vol.66(5),pp.223-250.Livemint(2023),“Kejriwal government announces hike in minimum monthly wages for workers in Delhi”,The Mint,19 October.Maillo,J.,Ramrez,S.,Triguero,I.

327、,&Herrera,F.(2017).kNN-IS:An Iterative Spark-based design of the k-Nearest Neighbors classifier for big data.Knowledge-Based Systems,117,3-15.Maillo,J.,Triguero,I.,&Herrera,F.(2015,August).A mapreduce-based k-nearest neighbor approach for big data classification.In 2015 IEEE Trustcom/BigDataSE/ISPA(

328、Vol.2,pp.167-172).IEEE.Mehta,S.,Shen,X.,Gou,J.,&Niu,D.(2018).A new nearest centroid neighbor classifier based on k local means using harmonic mean distance.Information,9(9),234.MFIN.(2023).MFIN Micrometer Data as on 30 September 2023.Retrieved from https:/mfinindia.org/Resources/micrometerNational S

329、ample Survey Office.(2019).Note on Sample Design and Estimation Procedure of NSS 77th Round.Reserve Bank of India.(2023).Reserve Bank of India Bulletin.Retrieved from https:/rbidocs.rbi.org.in/rdocs/Bulletin/PDFs/RBIBULLETINDECEMBER20238F42E694CD9E4937824D9FCCF408F7A5.PDFSong,Y.,Liang,J.,Lu,J.,&Zhao

330、,X.(2017).An efficient instance selection algorithm for k nearest neighbor regression.Neurocomputing,251,26-34.Van de Velden,M.,Iodice DEnza,A.,&Markos,A.(2019).Distance-based clustering of mixed data.Wiley Interdisciplinary Reviews:Computational Statistics,11(3),e1456.25References Abu Alfeilat,H.A.

331、,Hassanat,A.B.,Lasassmeh,O.,Tarawneh,A.S.,Alhasanat,M.B.,Eyal Salman,H.S.,&Prasath,V.S.(2019).Effects of distance measure choice on k-nearest neighbor classifier performance:a review.Big data,7(4),221-248.Bhattacharya,D.,&Ghosh,I.(2022).Must Shylock Receive his Pound of Flesh?An Ethical Inquiry into

332、 the Insolvency and Bankruptcy Regimes for Natural Persons.https:/ibbi.gov.in/uploads/whatsnew/b5fba368fbd5c5817333f95fbb0d48bb.pdf.Bhattacharya,D.,&Ananth,B.(2021).The Curious Case of Fresh Start Process:An inquiry into historical antecedents,international analogues,and domestic substitutes.https:/

333、ibbi.gov.in/uploads/whatsnew/1d8b31fc65f7ac6f09a973be8f12f868.pdfCarlos A.et al.(2019),“Bankruptcy,Discharge,and the Emergence of Debtor Rights in Eighteenth-Century England.”Enterprise&Society,Vol.20(2)pp.457-506.Chomboon,K.,Chujai,P.,Teerarassamee,P.,Kerdprasop,K.,&Kerdprasop,N.(2015,March).An empirical study of distance metrics for k-nearest neighbor algorithm.In Proceedings of the 3rd internat