Financial Development and the Underground Economy

Transcription

1 Fnancal Development and the Underground Economy Salvatore Capasso Unversty of Naples Parthenope and CSEF Tullo Jappell Unversty of Naples Federco II, CSEF and CEPR Revsed September 7, 2012 Abstract We provde a theoretcal and emprcal study of the relaton between fnancal development and the sze of the underground economy. In our theoretcal framework agents allocate nvestment between a low-return technology whch can be operated wth nternal funds, and a hgh-return technology whch requres external fnance. Frms can reduce the cost of fundng by dsclosng part or all of ther assets and pledgng them as collateral. The dsclosure decson, however, also nvolves hgher tax payments and reduces tax evason. We show that fnancal development (a reducton n the cost of external fnance) can reduce tax evason and the sze of the underground economy. We test the man mplcatons of the model usng Italan mcroeconomc data that allow us to construct a mcro-based ndex of the underground economy. In lne wth the model s predctons, we fnd that local fnancal development s assocated wth a smaller sze of the underground economy, controllng for the potental endogenety of fnancal development and other determnants of the underground economy. Key words: Underground Economy, Fnancal Development. JEL Classfcaton: G32, H26 Acknowledgements. We thank the Edtor, two anonymous referees, Salvatore Pccolo, Francesco Russo, semnar partcpants at the CSEF-IGIER 2012 Symposum on Economcs and Insttutons, and the Italan Mnstry of Unverstes and Research for fnancal support.

2 1. Introducton Recent estmates ndcate that the underground economy represents 10-15% of GDP n developed countres and 30-40% n developng countres. In some countres, such as Panama and Bolva, almost 70% of GDP s hdden (Schneder, 2007). Apart from ethcal and poltcal concerns, a large share of underground economy s a serous ssue for governments and polcy makers snce t dstorts nvestments, exacerbates ncome nequalty, and hampers growth. 1 Many factors explan the emergence and sze of nformal actvtes. A hgh level of taxaton, a cumbersome legslaton, hgh payroll taxes and labor costs are only some of the many factors whch may push frms nto nformalty. Among these factors, the avalablty of credt and ts cost have receved lttle attenton. In ths paper we study how the choce to operate underground (and to what extent) nteracts wth fnancal development. As n Ellul et al. (2012), the startng pont of our analyss s that the ablty to reveal and sgnal revenues reduces nformaton frctons and the cost of credt. When frms or ndvduals operate underground ther ablty to sgnal revenues and assets s lower, and the cost of credt hgher. As fnancal markets develop, more effcent ntermedares enter the market and the cost of credt falls, ncreasng the opportunty cost of contnung to operate underground. In short, fnancal market development s negatvely correlated wth the sze of the underground economy. To clarfy our arguments, we propose a smple theoretcal model n whch agents choose between a low-return technology and a more advanced and rewardng technology. Investng n the low-return technology does not requre a loan, whle the hgh-return technology requres external fundng. We post that frms can reduce the cost of credt by pledgng more collateral, as n Jappell et al. (2005). Snce contracts are not completely enforceable, part of the pledged resources can be lost n the case of a dspute, for example because of judcal costs and neffcences. Pledgng more collateral, however, s costly because frms must dsclose ther revenues and assets to fnancal ntermedares and also to tax offcals. Hence, agents choose how much to nvest n the two technologes by tradng off the reduced fnancal cost of supplyng more collateral aganst the beneft of hdng revenues and operatng wth the low-return technology. The choce between the two technologes therefore s also a choce between the underground and the offcal economy. Fnancal development reduces the cost of credt and the 1 The underground economy encompasses many actvtes. Many are legal, many others are crmnal and llegal. The extent and varety of these actvtes s vast. In ths paper we refer only to actvtes that per se are legal, but whch are hdden to offcal statstcs and authortes. We use the terms underground, nformal, unoffcal more or less synonymously. 1

3 ncentves to operate underground, whle makng t more proftable to reveal the revenues from hgh-tech projects. Our model adds two mportant nsghts to exstng works. Frst, we take explct account of the technologcal choce that s nvolved when entrepreneurs choose to operate underground. There s compellng evdence that the underground economy thrves n mature and noncompettve sectors, and that underground frms do not nnovate, operate on a small scale, and mplement low-return technologes. 2 In our model the choce to operate n the underground economy s drven by technologcal reasons, and the model mples that hgh-tech frms operate manly n the formal economy, whle low-tech frms operate manly underground. Second, n our model agents can operate smultaneously n both sectors, because they choose the optmal levels of ncome and assets to dsclose to the tax authortes. Ths s n lne wth emprcal evdence showng that frms and ndvduals are seldom completely underground or completely transparent (Johnson et al. 2000). In the second part of the paper we challenge the model s predctons wth emprcal evdence, explotng the varablty n local fnancal development across Italan regons. The data are drawn from the Bank of Italy s Survey of Households Income and Wealth (SHIW) to buld an ndex of the underground economy based on ndvdual-level data. The ndex measures the level of work rregularty among Italan workers from 1989 to 2006, and ranges from 0 (actvty s only n the formal sector) to 1 (actvty s completely hdden). We regress ths ndex on an ndcator of fnancal development and other ndvdual and regonal varables. The results show that the underground economy s strongly negatvely correlated wth fnancal development. We fnd also that more compettve and nnovatve sectors dsplay lower levels of underground actvty. Most mportantly, n our emprcal approach we control for the endogenety of fnancal development usng the ndcator proposed by Guso et al. (2004). The paper s organzed as follows. Secton 2 revews recent lterature on the underground economy. Secton 3 presents the theoretcal model. Secton 4 descrbes our ndcator of job rregularty. Descrptve analyss and emprcal estmates are presented n Sectons 5 and 6, respectvely. Secton 7 concludes. 2 See Loyaza (1996), Batra et al. (2003) and Farrel (2004). 2

4 2. Determnants of the underground economy Because of the heavy burden on the economy, many studes have examned causes and consequences of underground actvtes. It s not easy to provde n-depth and exhaustve explanatons for why frms and ndvduals evade taxes or operate rregularly and underground. Hgh levels of taxaton, cumbersome legslaton and a tght regulatory system, often consdered to be the man determnants of underground actvtes, provde only partal explanatons. 3 Other factors play a role n shapng the underground sector, and among them the role of nsttutons s lkely to be the most relevant. 4 Indeed, nsttutonal falure such as poor contract enforcement, judcal neffcency, complex and arbtrary regulaton reduce the ncentve for frms and ndvduals to reveal ther revenues. In a recent work Schneder (2010) fnds that the underground economy s rooted n a combnaton of factors such as a large burden of taxaton and socal securty payments, strngent labor market regulaton, poor qualty of state nsttutons, and poor tax morale. The nsttutonal settng can sgnfcantly affect the choce of nformalty because the effcency of publc nsttutons and the qualty of publc goods provson are mportant determnants of the opportunty cost to operate underground. For ths reason lack of democratc partcpaton, low level of tax morale, nsttutonal dstrust are all factors whch affect postvely the sze of underground economy, see Dreher and Schneder (2010), Teobaldell (2011), Cerquet and Copper (2011). These factors play a major role, and mprovements n the qualty of nsttutons mght work much better n reducng the sze of the underground economy than other measures of deterrence, see Feld and Schneder (2010). Among the many nsttutons whch have been lnked to the underground economy, the degree of fnancal development has receved relatvely lttle attenton. Yet nformalty s assocated wth a hgher cost of credt, whch s an mportant component of the overall opportunty cost to operate underground. To the extent that fnancal development reduces the cost of credt, t ncreases the opportunty cost of nformalty. Some papers explore such relaton. Straub (2005) develops a model n whch frms choose between formalty and nformalty. Beng formal nvolves hgher entry costs but lower penaltes for defaultng and lower fnancal costs snce hdden ncomes cannot be used as collateral. In Antunes and Cavalcant (2007) entrepreneurs choose between a formal and an nformal sector by tradng off hgher entry costs and tax oblgatons n the formal sector aganst hgher fnancal costs n the nformal sector. Pant 3 See Fredman et al. (2000), Schneder and Enste (2000), Schneder (2005), Dabla Norrs et al. (2005; 2008). 4 See Loyaza (1996), Fredman et al. (2000), Johnson et al. (1998a; 1998b), Dreher et al. (2009). 3

5 et al. (2009) explore the relatonshp between employment, nformal actvtes and fnancal ntermedaton. The dea s that formal employment can spur fnancal ntermedaton snce workers wth regular jobs tend to use more ntensvely the bankng system as depostors. In Blackburn et al. (2012) entrepreneurs need external resources for nvestment and can reduce the level of nformaton costs and the fnancal outlays by supplyng more collateral. Supplyng more collateral, however, nvolves a hgher tax burden. Gven the fnancal costs, entrepreneurs choose whether or not to evade taxes and to operate underground. Ellul et al. (2012) suggest that when frms choose accountng transparency, they trade off the benefts of access to more abundant and cheaper captal aganst the cost of a hgher tax burden, and study ths trade-off n a model wth dstortonary taxes and endogenous ratonng of external fnance. On the emprcal front, some papers study the relaton between the underground economy and fnancal constrants. Dabla-Norrs et al. (2008) use a survey of regstered frms n 41 countres and fnd that fnancal constrants tend to nduce nformalty among small frms but not among large ones. Beck et al. (2010) fnd that access to fnance has a stronger mpact on tax evason for small frms, frms located n small ctes, and frms n ndustres that rely more heavly on external fnance. La Porta and Shlefer (2008) fnd that the underground economy s negatvely assocated wth the avalablty of prvate credt and ndvduals subjectve assessment of ther access to credt. Usng cross-country data, Bose et al. (2008) fnd that bank development s negatvely assocated wth the sze of the underground economy. Ellul et al. (2012) use mcroeconomc data from Worldscope and from the World Bank Enterprse Survey and fnd that nvestment and access to fnance are postvely correlated wth accountng transparency and negatvely wth tax pressure. They also fnd that transparency s negatvely correlated wth tax pressure, partcularly n sectors where frms are less dependent on external fnance, and that fnancal development encourages greater transparency by frms that are more dependent on external fnance. Exstng emprcal studes, however, do not address the ssue of endogenety and the potental reverse causalty argument that a large underground economy lmts the growth of fnancal ntermedares. 5 5 Gatt and Honorat (2008) use Italan regonal data and fnd that fnancal development s negatvely affected by ndcators of the underground economy. 4

6 3. The Model We consder an economy wth a large number of banks whch lend to a contnuum of rsk neutral entrepreneurs, denoted by. Banks have a postve and exogenous cost of ssung a unt of loan, R + δ R %, whch s the sum of the cost of rasng funds, R, and an ntermedaton cost δ. Each entrepreneur s endowed wth an llqud asset, A, whch s unformly dstrbuted n the nterval [0, A ]. The asset (or part of t) can be used as loan collateral. We denote the fracton of A dsclosed to the bank and employed as collateral as γ, wth γ [0,1]. Hence, banks observe γ A but not γ or A separately. The fracton of the asset that s hdden, (1-γ )A, s not observed by any other agent or hence by government. Each entrepreneur can undertake two types of nvestment, Hgh-Tech and Low-Tech projects (HT and LT, respectvely). HT projects are rsky, requre a loan and operate under a technology wth constant returns to scale. LT projects do not requre a loan but operate wth a less rewardng, decreasng returns technology. The assumpton that frms use smultaneously dfferent technologes wth dfferent returns s not a new one. Learnng costs, fnancal costs and other constrants may hamper the adopton of new technologes even when they are readly avalable. Snce Mansfeld s (1963) semnal contrbuton, economsts have attempted to study not only the dynamcs of nter-frm rates of dffuson (technology dffuson between frms) but also ntra-frm rates of adopton of new technologes (technology dffuson wthn a frm). In the presence of frctons and constrants frms may use for a long perod of tme dfferent technologes and tend to substtute old for new ones slowly. 6 Non-monotoncty n the dynamcs of adopton of new technologes mples that wthn the same ndustry, and n a partcular frm, more advanced and mature technology may coexst. Actually, the dynamcs of adopton tself may affect the returns of technologes (Arthur, 1989). Hgh rates of adopton lead to nnovaton and further mprovements. The more the technologes are adopted, the more knowledge s ganed from ther use and the more they are mproved upon, a process that Rosenberg (1982) descrbes as learnng by usng. More competton between technologes can enhance ths process, whch s the reason why more dynamc and more compettve sectors tend to nvolve a prevalence of 6 Mansfeld (1963) provdes the example of the desel locomotve whch, n the nterwar perod, substtuted the power steam slowly. To gve a more recent example, Hollensten and Woerter (2008) analyze the case of E- commerce as a technology whch coexsts wth tradtonal channels of trade. Recent work provdes supportng evdence of slow technology dffuson, even wthn partcular frms (Battst and Stoneman 2003, 2005; Capasso and Mavrotas, 2010). 5

7 hgh returns technologes. The opposte apples to mature and stagnant technologes where lack of nnovaton and ncreasng costs lead to decreasng returns. Followng these arguments, we assume that LT projects operate n the underground economy, whle HT projects operate n the formal sector. Indeed, we show that nvestment n LT projects nvolves tax evason, whle nvestment n HT projects requres entrepreneurs to reveal ther revenues. The match between LT and HT projects and the formalty of the economy accords also wth the dea that operatons n the underground economy rely on self-fnancng and more tradtonal projects. Frms engaged n the formal sector, n contrast, rely more heavly on external fnance and mplement more technologcally advanced projects. In the remander of ths secton we study the condtons under whch entrepreneurs operate n the formal sector, n the underground economy, or n both. Next, we study how fnancal development affects these decsons and the level of nvestment The two projects We assume that the LT project does not requre a loan, and that t can be carred out usng the llqud asset A to purchase Low-Tech captal K LT. If entrepreneurs undertake an LT project they operate wth a decreasng returns to scale technology, accordng to the followng producton functon: Q LT = Φ (1) K α LT LT projects are completely hdden to both lenders and government. Entrepreneurs nvest n these projects the share of the llqud asset whch s not pledged as collateral. Hence, f γ A s the fracton of the asset dsclosed to the bank n order to obtan a loan to fnance the HT project, the captal nvested n the LT project s K LT =(1-γ )A. HT projects operate under constant returns to scale. They requre a loan L and delver Q HT =QL unts of output wth probablty p and 0 unt of output wth probablty (1-p). Each HT project has a postve net present value: pql > RL % 6

8 There s no nformaton asymmetry between borrowers and lenders, and banks can always observe whether projects succeed or fal. However, as n Jappell et al. (2005), we assume that only part of the proceeds of the nvestment can be pledged aganst the loan. In partcular, we assume that n case of success lenders can recover at most a fracton θ of output (QL ), and a fracton φ of the collateral, wth θ [ 0,1] and φ [ 0,1]. The remanng fracton of output (1-θ) and collateral (1-φ) can be nterpreted as the amount of resources requred by the judcal system for ts functonng. One can thnk of ths loss as the cost of premature lqudaton of the nvestment or, alternatvely, as the cost of judcal effcency. 7 Thus, n the case that the project succeeds lenders obtan θql + ϕγ A unts of output, whle n the case of falure they obtan ϕγ A. We denote by R R % the agreed repayment per unt of loan. Ths repayment s set after borrowers supply the collateral γ A. In a compettve credt market, banks expected profts are zero and hence: RL % = pr L + (1 p) mn[ R L, ϕγ A ] (2) Dependng on the amount of collateral, the zero proft condton (2) determnes three possble cases. A frst case (Case A) arses f the collateral s suffcent to repay the lender f the project should fal, that s ϕγ A R L. From equaton (2) t s clear that the requred nterest rate s equal to the lowest possble rate; that s, the bank s cost of supplyng the loan s: R = R % (3) Only borrowers wth large endowments can access ths contract. Recallng that γ [0,1] and that the condtonϕγ A R L must be satsfed, to access ths contract the collateral requred s A A, wth A RL % / ϕ. Max Max A second case (Case B) arses f the collateral would be nsuffcent to repay the lender were the project to fal ( ϕγ A < R L ). Usng equaton (2) t s straghtforward to show that the requred nterest rate s now: 7 I.e., f borrowers dspute the clam, lenders can brng the case to court and recover a fracton of the output and collateral. 7

9 R% 1 p γ A R = ϕ p p L In ths case the nterest rate s a decreasng functon of the pledged collateral, and greater than n Case A. The thrd case arses f the amount of the collateral s nsuffcent to repay the lender even were the project to succeed (Case C). Ths occurs f the collateral s nsuffcent to cover the bank s cost of fundng. Let us denote by A mn the level of the endowment, A, below whch the expected return on the project does not cover the cost of fundng: A RL % ϕ mn = pθ QL ϕ In ths case, potental borrowers wth endowments A mn (4) (5) < A are excluded from credt (whle borrowers wth A A can access the fnancal contract as n Case B). For smplcty, we rule mn out Case A and focus on a stuaton n whch A < A, that s, no borrower has enough collateral Max to fnance a HT project at the nterest rate R %. Thus, we assume that, regardless of the dsclosed collateral γ [0,1], all borrowers are fnancally constraned. The problem of fnancally constraned borrowers s to choose the optmal level of the ntal asset dsclosed to the bank (γ A ). Ths choce nvolves a trade-off. The hgher the level of the pledged collateral γ A, the lower wll be the cost of the loan (see equaton(4)) and, n turn, the return on the HT project. However, by dsclosng the asset, borrowers face two costs: a drect cost due to hgher taxaton, and a hgher opportunty cost due to the ncome loss n operatng the LT project on a smaller scale The dsclosure choce The optmal share of dsclosed collateral, γ, depends on borrower s expected utlty, whch, n turn, depends on the avalable fnancal contract. We know from the dscusson n the prevous secton that by pledgng a suffcent level of collateral, γ A A, borrowers can obtan a loan (under the fnancal contract of Case B) and run the HT project. The remanng (and hdden) part of the asset can be alternatvely nvested n the LT project. Therefore, the optmal choce of collateral ultmately s a choce between the HT and LT projects. The mplcaton s that f γ =1 mn 8

10 only HT projects are undertaken, f γ =0 that only LT projects are operated, and ntermedate values of γ ndcate nvestment n both projects. We now determne the return on each project and each borrower s expected utlty. We denote by t the tax rate leved on the dsclosed asset and on the revenues generated by the HT project. The expected utlty from runnng the HT project s: E ( u) = (1 t) p[( Q R ) L + γ A ] (6) HT The expresson states that the project s successful wth probablty p, and delvers ( Q R ) L unts of output. Revenues and the dsclosed asset are observed by government and taxed at the rate t. In the opposte case, wth probablty 1 p, the project fals and produces no revenues. Snce the nterest rate s gven by equaton (4), the expected utlty from runnng the HT project can be rewrtten as: E ( u) = (1 t) p[( pq R % ) L + (1 p) ϕγ A + pγ A ] (7) HT The fracton of the asset not dsclosed as collateral, K = (1 γ ) A, s nvested n the LT LT project. Snce the revenues from LT projects are not observed by government, no taxes are pad, and the expected utlty s smply: E ( u) = Φ[(1 γ ) A ] α (8) LT Combnng equatons (7) and (8), we obtan the borrower s expected utlty: E( u) = E ( u) + E ( u) = HT LT = (1 t) p[( pq R % ) L + (1 p) ϕγ A + pγ A ] + Φ[(1 γ ) A ] α (9) The optmal choce s determned maxmzng equaton (9) wth respect to γ. The frst order condton of the problem s: αφ (1 γ ) = Ω A 1 1 α, (10) 9

11 assets. 8 The model shows that the optmal choce of collateral and, correspondngly, the extent to where Ω = (1 t)[(1 p) ϕ + p]. Ths condton defnes the optmal γ as a functon of A,.e. γ = γ ( A). It s straghtforward to verfy that, other thngs equal, a hgher collateral ncreases γ dsclosure,.e. > 0. The result depends on the characterstcs of the two projects. Borrowers A choose how much to dsclose (γ A ), equatng the margnal returns from the HT and LT projects. The LT project dsplays decreasng returns and, hence, there s a unque level of captal K = (1 γ ) A that maxmzes the project s return. Ths mples that borrowers wth hgher LT levels of asset endowment, A, wll maxmze returns by nvestng a lower share of the asset, γ, n the LT project and, correspondngly, a hgher share n the HT project whose return s a lnear functon of the collateral. Hence, t could be argued that dsclosure ncreases wth the level of whch borrowers nvest n the LT project and hde ther ncome, depends on the relatve returns from the two projects. In the next secton we study how fnancal market development (a reducton n the cost of credt) affects the relatve return and the sze of the underground economy. We focus below on the effects of mprovements n judcal effcency, tax reforms, and changes n the technology of underground actvtes Fnancal development Fnancal development s a multfaceted phenomenon. It nvolves typcally the emergence of new and thcker captal markets, the ntroducton of new fnancal nstruments, and greater competton between ntermedares. Yet, n general, t s possble to argue that fnancal development entals a lower cost of rasng funds. In our model we consder fnancal development as correspondng to a smaller ntermedaton cost δ and a lower cost of fnance, R + δ R %. In ths framework dsclosure entals a trade-off. Dsclosng collateral reduces the cost of accessng external fundng, but ncreases the tax burden. Furthermore, once the collateral s 8 To explan ths result, recall that A = K LT + K, where HT K HT γ A. Snce the return on the LT project s maxmzed at a gven level of nvested captal, ˆ (1 ) [ / ] 1/(1 α K A γ = αφ Ω ), any further ncrease n A wll be nvested n the HT project. Hence, gven K LT LT = Kˆ, any ncrease n A wll rase K HT and mply a hgher γ. LT 10

12 dsclosed, t cannot be used n the LT sector, whch reduces revenues from LT projects. Fnancal development reduces the sze of the underground economy only f t relaxes the credt constrants, nducng more agents to borrow. Therefore agents wth very low-endowment A < A ) are not affected by fnancal development. To see ths, recall that ther expected utlty ( mn s: E( u) = E ( u) = Φ[(1 γ ) A ] α. (11) LT The above expresson mples that these entrepreneurs set γ = 0. Hence, any change n the cost of credt does not affect ther nvestment decson. Instead, entrepreneurs whose assets are above A are able to access the credt market, set γ < 1 and run both projects. Moreover, for mn these entrepreneurs the choce of collateral s a monotonc and ncreasng functon of ther endowment, that s, the hgher A, the hgher γ. Fgure 1 shows that the dsclosure functon γ = γ ( A ) s a step functon. For 0 A < A, the functon concdes wth the horzontal axs. mn For Amn A A, the functon s determned by equaton (10), and s therefore concave. 9 Snce we assume that A s dstrbuted unformly over the nterval 0, A, the area below the dsclosure functon measures total dsclosed assets. 10 Let us now see how fnancal development affects underground actvtes. We know from equaton (5) that a reducton n R % reduces the threshold level of collateral Amn whch allows borrowers to access credt. Fgure 2 shows that also a reducton n R % reduces A mn to ts new value A. Borrowers wth A' mn A < A who prevously were credt constraned and mn ' mn operated only n the underground economy, now dsclose part of ther asset, obtan a loan, and run the HT project. The dsclosure functon γ = γ ( A ) shfts accordngly: the concave porton of the curve shfts to the rght, whle the secton lyng on the horzontal axs shrnks, as does the actvty n the formal economy. We summarze ths result n the followng proposton. 9 2 γ γ Dfferentaton of equaton (10) shows that > 0 and > 0 2 A ( A ) 10 Entrepreneurs wth assets just above A mn have two optons. One s to set γ hgher than the optmal value as determned by equaton (10) n order to supply enough collateral to access credt,.e. γ A A. The alternatve s to set γ = 0. The choce of these entrepreneurs ultmately depends on the relatve return of the two technologes. It s possble to show that there exsts a threshold level of asset mn  > A below whch these entrepreneurs wll optmally set γ = 0, and that above such threshold ther choce s dctated by equaton (10). Redefnng the threshold, however, does not change the shape of the dsclosure functon. Furthermore, snce the threshold s a functon of A mn, to smplfy exposton we focus on A mn. mn 11

13 Proposton 1: Other thngs equal, fnancal development ncreases the opportunty cost of tax evason, lowers underground actvty, reduces credt ratonng, and stmulates nvestment n new technologes. Note that n our settng fnancal development also mples technologcal mprovement; that s, more frms operate HT projects. Ths s n lne wth the emprcal evdence showng that a reducton n the sze of the underground economy s assocated wth more effcent use of resources and allocaton of nvestments, see e.g. Loyaza (1996) and Farrel (2004). A second mplcaton of the model s that fnancal market development reduces credt ratonng, but can never elmnate t even n the best scenaro. Ths depends on our assumpton that LT projects operate wth decreasng returns. For low levels of assets - and, hence, for low levels of K LT - LT projects always domnate HT projects. Usng equaton (10) t s easy to dentfy the mnmum level of asset, A LT, below whch LT projects are always preferred to HT projects: A LT 1 αφ 1 α = Ω As shown n Fgure 2 all agents wth 0 A ALT choose γ = 0 and run only the LT project, regardless of the cost of fnancal ntermedaton. Thus, fnancal development can reduce underground actvty only f A > A. The model also mples that the effect of fnancal LT development on the sze of the underground economy s non monotonc and that t s stronger at low levels of fnancal development. Ths s because of the concavty of γ ( A ), whch measures the amount of dsclosed asset. (12) 3.4. Judcal effcency As wth the cost of fnancal ntermedaton, any other factor that affects the relatve returns of the two projects also affects the choce of collateral and, from equaton (10), the choce of γ. Ths mples that an ncrease n the tax rate t reduces the expected return from HT projects and the optmal γ. For the same reasons, an ncrease n the productvty of LT projects (an ncrease n Φ) rases the proftablty of the project and reduces γ. In graphcal terms, as t ncreases and the expected return of HT projects falls, the dsclosure functon γ ( A ) shfts downwards. Ths mples that each entrepreneur wll dsclose a lower share of assets as collateral. 12

14 Notce that n our model, taxaton does not affect credt ratonng because the tax rate does not enter equaton (5) and therefore the value of A mn. The model also suggests that changes n judcal effcency may affect γ. To see ths, recall that we nterpret the terms (1 θ ) Q and HT (1 ϕ ) A as the amount of resources lost n the case of a legal dspute, and that an ncrease n θ or φ sgnals a more effcent judcal system. 11 These two parameters affect γ n two ways: () by reducng credt ratonng, and () by changng the relatve return between HT and LT projects. The frst channel operates because a better judcal system (an ncrease n θ or φ) reduces the threshold A mn (see equaton (5)) and the regon of credt ratonng. Dsclosed assets ncrease accordngly, and the underground economy shrnks. An ncrease n φ also rases the return on HT relatve to LT projects. Ths ncreases also the ncentve to dsclose assets and to nvest n the HT technology. 12 Fgure 3 shows how an mprovement n judcal effcency affects γ. The ncrease n θ or φ (the frst channel) reduces A mn to the new value (the second channel) shfts the γ ( A ) functon s mn A ' mn. The ncrease n the return of HT projects functon upwards. Hence, the new support of the dsclosure [ A', A ]. The sze of the underground economy shrnks because, n the new equlbrum, entrepreneurs who prevously were recevng credt, borrow more and dsclose more assets, whle those who prevously were credt ratoned obtan loans and dsclose part of ther assets. Judcal effcency also amplfes the mpact of fnancal development on the sze of the underground economy. Fgure 3 llustrates also the nteracton between fnancal development and judcal effcency. An mprovement n judcal effcency shfts the dsclosure functon upwards, from γ ( A ) to the new value γ ( ) 1 A. The thresholds Amn values and A LT decrease to the new A' and mn A ' LT. Hence, fnancal development (a reducton of A to mn A ' mn ) has a larger mpact on the underground economy when the judcal system s effcent. The reason s that when the courts are effcent, fnancal development nduces entrepreneurs to dsclose a larger fracton of ther assets, so that the underground economy decreases by a larger amount. We summarze the results of ths paragraph n the followng proposton: 11 In keepng wth the model smplcty, we assume that t s costless to ncrease judcal effcency. In general, rasng judcal effcency mght requre publc resources and therefore a hgher tax rate. However, some reforms mght ncrease the productvty of the judcal sectors even at the same level of expendtures. For nstance, Covello et al. (2010) show theoretcally and emprcally that task jugglng (the spreadng of effort across too many actve trals) decreases the performance of Italan judges, rasng the chances of long duraton of trals and explodng backlogs. Better management of judcal dstrcts (.e., a reducton n task jugglng) would ncrease the productvty of judges wthout rasng publc expendtures. 12 Ths result can be verfed by nspectng the frst order condton (10). 13

15 Proposton 2: Other thngs equal, an mprovement n judcal effcency reduces the sze of the underground economy. Judcal effcency also amplfes the mpact of fnancal development on the sze of underground economy The technology gap Emprcal evdence shows that the sze of the underground economy dffers consderably across sectors, see Johnson et al. (2000), Batra et al. (2003) and Farrel (2004). For nstance, n the constructon ndustry underground actvtes are wdespread, whle the chemcals and drugs sectors are comprsed mostly of formal enterprses. One of the reasons for ths s due to labor market regulaton, but most of the dfference depends on the technologes nvolved n these sectors. As we argue above, optmal nvestment and dsclosure polces depend on relatve returns (and ther determnants) from the avalable technologes. The frst order condton (equaton (10)) shows that a decrease n the return of LT projects (Φ) ncreases dsclosure at each level of A. On the other hand, a hgher return of hgh tech projects (Q) does not affect nvestment n HT projects drectly, but reduces credt ratonng by lowerng A mn (see equaton (5)) and reducng the sze of the underground economy. Therefore the sze of the underground economy n each sector depends on the relatve returns of nvestment projects and the degree of credt ratonng. More dynamc and compettve sectors (e.g. the fnancal sector, or the chemcals ndustry) tend to have hgher returns (Q) from ther HT projects. Frms n these sectors tend to have lower rates of underground actvtes because they are less lkely to be credt constraned. These sectors are more compettve, more technologcally advanced, and are lkely to exhbt a lower technologcal gap between HT and LT technologes. The opposte happens n less dynamc sectors (e.g. constructon or retal), where new technologes are ntroduced at slower rates and frms can survve despte the mplementaton of mature technologes. In our model, gven the return from HT projects, the parameter Φ measures the technologcal gap between the two projects. A lower Φ ndcates a larger gap and therefore s typcal of less dynamc (backward) sectors whle more dynamc (advanced) sectors feature a hgher Φ. As shown n Fgure 4, our model predcts that the mpact of fnancal development s larger for backward sectors. Snce Φ s lower, the dsclosure functon of the backward sector, 14

16 B A γ ( ), les above the dsclosure functon of the advanced sector, γ ( ). For the same reason, A B credt ratonng n the backward sector s larger (whch features A mn A ( A mn A ) than n the advanced sector ). Ths mples that n backward sectors the mpact of fnancal development on the sze of the underground economy s relatvely stronger. We summarze the dscusson n ths paragraph n the followng proposton: Proposton 3: The sze of the underground economy depends on the technologcal gap between LT and HT projects. More mature and less dynamc sectors tend to dsplay hgher rates of underground actvtes. Other thngs equal, n these sectors the mpact of fnancal development on the underground economy s larger. 4. The data To test the man mplcatons of the model we use the Bank of Italy s SHIW, whch allows us to construct an ndex of underground actvtes based on mcroeconomc nformaton. SHIW s a bannual cross-secton of about 8,000 households and 24,000 ndvduals, and provdes detaled nformaton on demographc varables, ncome, consumpton, and wealth. Survey data are avalable from 1977, but the man varable of nterest for ths paper s avalable only n 1995, 1998, 2000, 2002 and We exclude ndvduals who do not report years of contrbutons, are not part of the labor force, or who work n the agrcultural or publc sectors. Our fnal sample ncludes 11,781 observatons. The SHIW s a representatve sample of the Italan resdent populaton. The sample desgn s smlar to the Labor Force Survey conducted by ISTAT (the Italan natonal statstcs agency). 13 Data are collected through personal ntervews. Questons concernng the whole household are addressed to the famly head or the person most knowledgeable about the famly fnances; questons about ndvdual ncomes are answered by ndvdual household members wherever possble. The unt of observaton s the famly, whch s defned to nclude all persons resdng n the same dwellng who are related by blood, marrage, or adopton. Indvduals selected as partners or other common-law relatonshps are also treated as famles. 13 Samplng s carred out n two stages: the frst covers the selecton of muncpaltes, the second the selecton of households. Muncpaltes are categorzed nto 51 strata, defned by 17 regons and 3 classes of populaton sze (over 40,000, 20,000-40,000, less than 20,000). All muncpaltes n the frst group are ncluded; those n the second and thrd groups are selected randomly wth a probablty proportonal to ther populaton sze. In the second stage households are selected randomly from regstry offce records. 15

17 For obvous reasons, tax evason and underground actvtes are dffcult to detect and measure. Indvduals and frms who evade taxes or operate rregularly tend to hde ther ncome from the government, and hence, are unlkely to release nformaton on ther hdden actvtes. Ths makes t dffcult to obtan drect data on underground actvtes and s the reason why economsts have tred dfferent ndrect measurement methods, such as the currency demand approach, the gap between effectve and potental electrcty consumpton, or the multple ndcators approach. 14 These methods are based on macroeconomc estmates of the sze of the underground economy, and have at least two lmtatons: () they are subject to large measurement errors; and () by constructon, the resultng ndcators of underground actvtes are strongly correlated wth other macroeconomc varables. We overcome some of these measurement problems by constructng an ndex of underground economy usng mcroeconomc data. Of course, our survey ncludes no drect questons about the extent to whch each ndvdual evades n taxes or works rregularly. However, we can nfer the degree of rregularty and evason through the followng two questons, whch are posed to each ndvdual ntervewed: (1) How old were you when you started workng? and For how many years, or months, dd you or your employer not pay, socal securty contrbutons? 15 From these two questons we can construct an ndex of rregular actvtes by dvdng the number of years not covered by socal securty contrbutons by the length of the workng lfe. There are several advantages to usng these questons. Frst, they are drectly related to evason of socal securty contrbutons and rregular work, among the man sgns of underground actvty. Second, whle respondents are unlkely to reply to drect questons about ther jobs, they may be more nclned to report ndrect nformaton on contrbutons towards ther pensons. Thrd, snce our objectve s to study the relaton between the underground economy and fnancal development, t s straghtforward to merge our ndex of rregular work wth the ndex of fnancal development proposed by Guso et al. (2004), whch s estmated usng the same data. Fnally, and most mportantly, our analyss explots regonal varablty n the level of fnancal development n a sngle country. By focusng on the same jursdcton, we overcome the problem 14 Accordng to ths approach, a country s shadow economy s treated as a latent varable whch s then mputed usng several ndcators and cause varables. Ths method provdes the wdest country coverage and therefore s used extensvely n the macroeconomc lterature, see Djankov et al. (2002), Loyayza et al., (2005) and La Porta and Shlefer (2008). 15 The socal securty contrbuton rate s 33% of the gross wage for prvate and publc employees and 20% n the case of self-employment. 16

18 that a relaton between underground economy and fnancal development arses because both varables are correlated wth other nsttutonal and macroeconomc ndcators. However, our ndcator also has some drawbacks. As wth many mcroeconomc varables, an obvous source of concern s msreportng and recall bas. Another concern s that years not covered by socal contrbuton mght be years of unemployment rather than years of rregular work. For ths reason, n our estmates we control for the local unemployment rate and per capta GDP at the provncal level. We also construct an alternatve measure of underground economy. Followng a standard approach of the lterature, we proxy underground actvtes by calculatng the fracton of ncome receved n cash. The dea s that nformal actvtes gve rse to cash transactons. As our frst ndcator, ths varable s based on the followng queston avalable n the SHIW: Last year, dd you receve part of your (or your famly) ncome n cash? and n whch fracton? As other proxes, ths ndcator has some lmtatons. One lmtaton s that t mght be assocated wth dfferent payment technologes, whch may themselves be related to the level of local fnancal development. For ths reason we use ths varable only as a robustness check. As already mentoned, the SHIW provdes also an ndcator of local fnancal development. Ths ndcator, proposed by Guso et al. (2004), measures the probablty that households have access to credt, that s, that they are not credt constraned. The SHIW asks households to report whether, n the 12 months before the ntervew, they have been dened credt or dd not apply for credt because they thought they would be turned down. Based on ths nformaton, and controllng for other relevant varables, Guso et al. (2004) estmate the probablty that a potental borrower s turned down for credt or dscouraged from borrowng, controllng for a wde range of ndvdual and regonal varables. The regonal dummes obtaned from the regresson model are then normalzed to be equal to zero n the regon wth the maxmum value of the coeffcent of the regonal dummy (Calabra s the least fnancally developed regon, whle March and Lgura the most developed), and therefore vares between zero and 1 (the hghest value s 0.58). Havng collected ndcators of rregular actvtes, fnancal development and judcal effcency, we can test some of the mplcatons of the model by estmatng equatons of the form: U = α X + α FD + α JUD + α Z + µ + ε (13) rs 1 rs 2 r 3 r 4 r s rs 17

19 where U s an ndcator of rregular actvtes for ndvdual n regon r and sector s, X a set of socoeconomc ndcators, FD the ndex of fnancal development, JUD a measure of judcal neffcency (duraton of trals, descrbed n Secton 6), Z a set of regonal ndcators, µ s sector fxed effects, and ε an error term. Proposton (1) suggests α 1 < 0, and proposton (2) α 2 > 0 (because longer length of trals s assocated wth less judcal effcency and an ncrease of underground actvtes). In the emprcal secton we also report separate estmates of equaton (13) by sector to verfy the mplcaton of proposton (3) that α 1 and α 2 should be larger n more mature sectors. Our analyss of the relaton between fnancal development and the underground economy needs to address the ssue of potental reverse causalty and endogenety of fnancal development FD n equaton (13). In partcular, an ncrease n underground actvtes (e.g. due to an ncrease n general taxaton) reduces the demand for credt, hamperng fnancal market growth. Smlarly, low GDP growth mght reduce the demand for loans and fnancal development, whle at the same tme ncreasng underground actvtes. Ths mples that smply observng that low fnancal development s assocated wth a hgh level of underground actvtes does not necessarly mean that low fnancal development actually causes more underground actvtes. In our mcroeconomc data, we address the endogenety of fnancal development relyng on an nstrument proposed by Guso et al. (2004) whch s correlated wth fnancal development, but s not affected by the degree of underground economy. The nstrument s based on the characterstcs of the 1936 Bankng Law, whch over tme has constraned the growth of the Italan bankng system and s an exogenous determnant of the trajectores of local fnancal development. Followng a perod of frequent bankng crses, n 1936 Italan legslators attempted to stablze the fnancal system by strctly lmtng n each regon the number of banks and bank branches. In achevng ths am, the law has worked very well, as wtnessed by the fact that the number of new branches n Italy after 1936 has expanded very lttle. Yet, n some regons and for some local credt nsttutons (such as savngs banks and cooperatve banks) the 1936 Bankng Law has been less constranng. Therefore the 1936 Law explans a large part of the varablty n local fnancal development even after 60 years. Guso et al. (2004) test ths hypothess by estmatng the correlaton between the ndex of regonal fnancal development and the characterstcs of the bankng system before the 1936 Law. They fnd that 1936 bank branches, local branches, savng banks and cooperatve banks (each n per capta terms) explan 72% of the regonal varaton n credt supply n the 1990s. In our emprcal estmates we use the same 18

20 nstruments whch s uncorrelated to underground economy to control for endogenety n fnancal development. 5. Descrptve analyss Our mcroeconomc ndcator of rregular actvtes s consstent wth macroeconomc estmates from dfferent sources. Fgure 5 plots the regonal averages of the ndex of job rregularty aganst a smlar ndex, produced by ISTAT but based on the Labor Force Survey. Despte the very dfferent methods of elctaton, a strong correlaton between the two measures s evdent (the correlaton coeffcent s 0.87). Both ndcators show that Southern regons feature the hghest levels of underground economy. In partcular, n Campana, Scly, Sardna, and other Southern regons the rregular job rate exceeds 30%, more than twce as hgh as the level of rregularty n Northern regons such as Frul and Emla-Romagna. The South s also much less developed n terms of per capta GDP, nfrastructure, and human captal. It s characterzed by more corrupton, less effcent government, and hgher levels of organzed crme. Each of these factors potentally contrbutes to generatng a large underground economy. Yet these regons also dsplay relatvely low levels of fnancal development, and we argue that ths channel plays an mportant role n shapng the underground economy. Fgure 6 plots the relaton between fnancal development and the sze of the underground economy. We use regonal averages for the perod The correlaton s strongly negatve (-0.81) and statstcally dfferent from zero at the 1% level. Fgure 6 shows a strong geographcal dvde. For example, n Campana a hgh rregular job rate (36%) s coupled wth an ndex of fnancal development of only 3%. In contrast, Lombardy (the rchest regon n the North) has a much lower rregular job rate (16%) but a much hgher ndex of fnancal development (43%). In Italy, as n many other countres, there are sgnfcant dfferences by sector n the level of underground actvty. The ndex of job rregularty reaches 30.8% n the constructon sector, 25% n the retal and toursm sectors, but s much lower (12% and 15% respectvely) n the fnancal and manufacturng sectors. These dfferences clearly reflect structural and technologcal dfferences between sectors. Underground actvtes are more wdespread n low value added sectors wth relatvely low competton and smaller frm szes. Note that ths s one of the predctons of the model, because frms operatng n more mature sectors have fewer ncentves to nvest and lower opportunty costs of hdng revenue. 19

21 Descrptve statstcs show also that the sze of the underground economy depends on the nature of employment. Self-employed, professonals, and entrepreneurs are much more lkely to work n the underground economy (the ndex of job rregularty n these occupatons ranges from 24% for self-employed to 26% for professonals and entrepreneurs). For managers (8%), and clerks (12%) rregular actvtes are much less wdespread. One reason for ths s that, n Italy, employers deduct the tax before transferrng wages to employees. Ths mples that t s much more dffcult for employees to evade taxes and socal contrbutons. Therefore hdden actvtes arse from extra work not supported by a formal employment contract. In the next secton we present regressons for the relaton between local fnancal development and the underground economy, controllng for possble sources of endogenety and other factors (such as sector and occupaton) whch mght nfluence the relatonshp. 6. Regresson evdence In our emprcal estmates we regress the rregular job rate on the ndcator of local fnancal development and a set of ndvdual varables (gender, age, years of educaton, martal status, dsposable ncome). Each regresson also ncludes tme dummes; some of the specfcatons nclude occupaton or/and sector dummes. We also nclude an ndcator to control for judcal neffcency, usng ISTAT data. Ths ndcator measures the length of ordnary cvl trals, that s, the tme elapsng from the date of the ntal recordng of a tral to the sentence, for actons requrng adjudcaton of substantve rghts concernng credt and commercal matters such as loans, sale of real estate or goods, rentals, negotable and quas-negotable nstruments, and nsurance. 16 The enforcement cost s drectly related to the length of the judcal process. A long tral ncreases the legal expenses and, for dsputed loans, the nterest ncome that s forgone when the collateral does not cover the judcal costs. Moreover, durng the tme of the tral, the credtor s exposed to the danger of asset substtuton by the debtor and to unexpected changes n the value of collateral. Therefore we expect that judcal neffcency s assocated wth more underground actvtes (α 2 >0 n equaton (13)). 16 A narrower classfcaton of legal acton (e.g., loans only) produces too few observatons for each dstrct-year cell to compute relable ndcators of judcal neffcency. For the same reason we do not consder the length of appeals n cvl cases and bankruptcy procedures. 20

22 We start our analyss by presentng the OLS regressons. Snce some of the rght-hand sde varables vary only between provnces or regons (judcal neffcency, local unemployment rate, fnancal market development), standard errors are adjusted for clusterng at the provncal level. Table 2, column 1 presents our baselne model. The demographc varables explan a substantal part of the varablty of the rregular job rate. In partcular, we fnd that women and younger ndvduals wth lower levels of educaton are more lkely to work rregularly, whle hgher dsposable ncome ncreases the lkelhood of operatng n the formal sector. The man varable of nterest s fnancal development. The coeffcent of ths varable s negatve and statstcally dfferent from zero at the 1% level, whch s consstent wth the model s predcton (α 1 <0 n equaton (13)). Its mpact s szable: rasng fnancal development by 10 percentage ponts (approxmately the dstance between Tuscany and Emla-Romagna) reduces the rregular job rate by 2.2 percentage ponts. Lower judcal effcency s assocated wth a hgher rate of rregular workng (the coeffcent s and s statstcally dfferent from zero at the 1% level). Note that the model n Secton 2 suggests that judcal effcency may affect the sze of the underground economy both drectly and ndrectly. Drectly, judcal effcency reduces the sze of the underground economy by ncreasng the opportunty cost of hdng ncome. Indrectly, an mprovement n judcal effcency ncreases the value of collateral and reduces the cost of credt. Therefore, the effect of judcal effcency s captured partly by the ndex of local fnancal development. In Table 2, column 2 we add to the baselne model a dummy for the South and an ndcator of socal captal (fracton of the populaton partcpatng n general electons n each provnce). 17 The coeffcents of both varables are postve, but only the dummy for the South s statstcally dfferent from zero. Snce Southern regons tend also to be the least fnancally developed and feature the hghest judcal neffcency, ntroducng ths dummy attenuates the mpact of fnancal development (coeffcent s ) and judcal neffcency (0.038). Furthermore, whle the coeffcent of fnancal development s stll statstcally dfferent from zero at the 1% level, the effect of judcal neffcency s now less precsely estmated than n the regresson n column 1. The thrd specfcaton n Table 2 repeats the estmaton ntroducng sector and occupaton dummes, and the results are essentally unaffected. The fnal specfcaton n Table 2, column 4 adds the provncal unemployment rate to control for the fact that some of the rregular work mght be due to spells of unemployment. The coeffcent of ths ndcator s not statstcally dfferent from zero, and agan the other coeffcents are unaffected. Other regressons wth 17 Other common proxes for socal captal (e.g., non proft organzatons) delver qualtatvely smlar results. 21