Microeconometric Analysis of the Determinants of Savings Bahaviour in Zimbabwe:

Transcription

1 Internatonal Journal of Busness and Management; Vol. 8, No. 10; 2013 ISSN E-ISSN Publshed by Canadan Center of Scence and Educaton Mcroeconometrc Analyss of the Determnants of Savngs Bahavour n Zmbabwe: Chkoko Laurne 1, Perre Le Roux 2 & Dzngra Canco 3 1 Department of Bankng and Fnance, Mdlands State Unversty, Gweru, Zmbabwe 2 Department of Economcs, Nelson Mandela Metropoltan Unversty, South Afrca 3 Department of Economcs, Mdlands State Unversty, Gweru, Zmbabwe Correspondence: Chkoko Laurne, Department of Bankng and Fnance, Mdlands State Unversty, Gweru, Zmbabwe. Tel: E-mal: chkokol@msu.ac.zw Receved: January 9, 2013 Accepted: March 6, 2013 Onlne Publshed: Aprl 27, 2013 do: /bm.v8n10p159 URL: Abstract A hgh savngs culture s a foundaton of a sustanable natonal socal, sound fnancal and economc polcy. The man obectve of ths paper was to gan nsghts on savngs behavour by ndvduals n Zmbabwe after the adopton of the multple currency exchange rate system. A mcro econometrc approach whch took nto account of ndvdual heterogenety was adopted. The selected ndvdual characterstcs nputted n the logt model as savngs predctors were age, martal status, relgon, educaton, poston n the household, household sze, type of accommodaton, place of accommodaton, employment status, ncome, number of people employed n the household, and expendture per month. The goal was to understand whch varables authortes and commercal banks mght need to target n order to promote a savngs culture n Zmbabwe after most people lost ther savngs n the banks due to hypernflaton. Secondly, we nvestgated wthn a sample of savers what determned the number of tmes ndvduals would depost money n the bank by use of Posson regresson model. Fnally, we came up wth a savngs predctng model whch helps banks dentfy targeted customers based on the ndvdual characterstcs. The results revealed that maorty of the varables were sgnfcant as predctors to savngs. However, ncome level proved to be the most sgnfcant varable as t had the least probablty value. To promote a savng culture, banks n Zmbabwe may need to consder brngng n dfferentated and talor made products for the hgh net worth and the mass market. Keywords: Zmbabwe, mcroeconometrc, savngs behavour, commercal bank, logt, posson 1. Introducton The government of Zmbabwe delberately allowed the use of a multple currency exchange rate system (MPC), whch was adopted on 30 January The multple currency system allows trade to be completed usng maor tradng currences, for example, the Unted States Dollar, Pound Sterlng, South Afrcan Rand, and the Botswana Pula. The new regme helped restore prce stablty and restart fnancal ntermedaton (MOF 2010, RBZ 2010). Nevertheless, t took a lot of effort to attract deposts back nto the system n Zmbabwe due to loss averson phoba after most people lost ther savngs n the banks due to hypernflaton. Gven ths background, t s mportant to understand how a savngs culture can be promoted n Zmbabwe. To acheve that, t s mportant to understand fundamentals that determne savngs. Aggregate natonal savng s composed of ndvdual, corporate and publc savngs. Compared wth developed world, Zmbabwe has no offcal household databank for each ndvdual household except tme seres data on natonal aggregate savng, mostly analysed usng macro-econometrcs models premsed on homogenety assumpton. Homogenety assumpton nduces error of measurement bas whch renders the parameters nconsstent and asymptotcally neffcent whch mght volate the Gauss-Markov assumptons. Post 2008 era, household savng was an ncreasngly crucal ssue for famles, ndvduals and ndustry resusctaton n Zmbabwe, gven that almost everythng was lost durng the pre 2008 era due to hypernflaton (RBZ, 2009; Mnstry of Fnance, 2009). Over the perod ndvdual depost rate averagng 1% was far below the average economc growth rate of 4% (RBZ, 2012). To make matters worse, n 2010, the household savngs wth fnancal nsttutons 159

2 Internatonal Journal of Busness and Management Vol. 8, No. 10; 2013 declned sharply by 10% (bd), ustfyng Beverly et al. s (2003) asserton that durng economc downturn and lqudty crss perods, ndvdual savngs and asset accumulaton are unversally vewed as desrable goals for economc turnaround stmul. Recent shfts n the structure of developng countres economes n general and Zmbabwe n partcular have brought debate on the possble solutons to such economc quagmres and, sgnfcance and determnants of household savng to the center of the polcy arena (Attanaso and Bank, 2001; Fsher and Montalto, 2011). Ths has stmulated a debate on the mportance and adequacy of household savngs for socal and economc growth fnancng through nvestment an necton n the crcular flow of ncome. Up to now, no consensus exst among economsts and polcy makers on the mcroeconomc determnants of observed household savng behavor and possble drvers of dfference n depost tmes wthn a month. A large body of lterature studes savngs from a macroeconomc perspectve. Unfortunately, macroeconomc approaches gnore the heterogenety n consumers whch realstcally reflect dfferent nfluences of savngs behavour by consumers. As a result, a mcro-econometrc approach s hghly recommended norder to have nsghts on savngs behavour by ndvduals n Zmbabwe after the adopton of the MPC. A mcro econometrc approach whch took nto account the ndvduals heterogenety was adopted. The obectve was to understand whch varables authortes and commercal banks would need to target n order to promote a savngs culture. Secondly, we nvestgated wthn a sample of savers what determned the number of tmes that an ndvdual depost money n the bank by use of Posson regresson model. Lastly, our obectve was to come up wth a savngs predctng model to be able to dentfy targeted customers based on the ndvdual characterstcs. 2. Lterature Revew Savngs are defned as a leakage out of the crcular flow of ncome meanng t s that part of ncome that has been left unconsumed and thus not gven back to the crcular flow of ncome but s rather spared for future use (Mankw 2007). The forgong of present consumpton for a hgher level of future consumpton s a better way of defnng savngs. Savngs also are done through the depostng of money n banks or other nvestment where a future greater return s expected thus savngs should equal nvestment wthout ths the purpose of savngs s dstorted. Revealng the mcro determnants of the aggregate savng rate s a paramount prerequste n desgnng a number of crucal polcy nterventons, from a tax and socal scrutny system to the layout of fnancal markets regulaton. Therefore t s not surprsng that the analyss of savng behavour has become one of the central ssues n emprcal mcro and macro-economcs (Deaton, 2005). Savngs are an nherent part to all players of the economy. Indvduals seek to smooth out consumpton over tme, savng n good tmes to consume n bad tmes. The precautonary motve for savngs fundamentally affects the savng behavour. Indvduals beneft from savng mostly for precautonary reasons, the fact that the future s uncertan means that acton should be taken before hand to hedge aganst any future adverstes death. The corporate world realses ts potental to boost the economy actvtes by accumulatng captal stocks through the converson of resources thus n dong so fnancal nsttutons and other frms hedge themselves aganst uncertantes lke durng a tme of economc downfall, savngs can also be used for the payng out of dvdends and thus ncreasng the welfare of shareholders. The government also needs savngs so as to fnance the government s publc sector borrowng requrement (.e. the funds n excess of taxaton that the government requres to fund ts actvtes). It s from natonal savngs that the government s able to mprove the welfare of ts ctzens e.g. the grantng of penson funds, drought relef ncentves and etc. Researches recognse savng as a crucal factor for the economc development as t enables converson of resources nto captal. Strong savng performance s crucal for macroeconomc balance. Two of the most well-known theores of savngs are the Lfe Cycle Hypothess and the Permanent Income Hypothess. 2.1 The Lfe Cycle Hypothess Lfe Cycle theory appears frst n a paper by Modglan and Brumberg (1954). They worked out a theory of spendng based on the dea that people make ntellgent choces about how much they want to spend at each age, constraned only by the resources avalable over ther lfe tme. For each ndvdual, t s assumed that an ncrease n lfe-tme resources leads to a proportonate ncrease n consumpton n all perods of lfe. As a result, consumpton s proportonal to lfetme resources or, what s more or less the same thng, to average ncome over the lfe tme (Deaton, 2005). One of the man mplcatons of the Lfe Cycle model s that ndvduals can separate consumpton from ncome, that s, consumpton s not affected by the tmng of ncome. Indvduals tend to dssave when they are young, because of relatvely low levels of ncome and hgh expendtures due to household formaton whle they tend to save n ther mddle years, wth asset holdngs reachng ther maxmum at retrement age, an ndcaton of non-lnear relatonshp between savng and age. Indvduals agan dssave 160

3 Internatonal Journal of Busness and Management Vol. 8, No. 10; 2013 durng retrement by usng ther accumulated assets, whch are entrely exhausted at death. Ths ustfy the ncluson of age and age-squared n the model. The Lfe Cycle theory leads to mportant and more advanced predctons about the economy as a whole, for example that natonal savng depends on the rate of growth of natonal ncome, not ts level, and that the level of wealth n the economy bears a smple relaton to the length of the retrement span. Further, t s the Lfe Cycle Hypothess that helps us thnk about a number of mportant polcy questons about whch we would otherwse have very lttle to say, for example how socetes should collectvely make provson for the ncreasng numbers of elderly and how does government provson nteract wth prvate provson (Deaton, 2005). Whle there have been many challenges to the theory of consumpton over the years, the Lfe Cycle Hypothess remans an essental part of economsts toolbox because t helps us to thnk about thngs such as the prvate and publc provson of socal securty, the effects of demographc change on natonal savng and the role of savng n economc growth. It s sometmes questoned whether the Lfe Cycle Hypothess of savngs s stll supported emprcally. In many aspects t has been reected. 2.2 The Permanent Income Hypothess Dynan et al. (2004) asserts that, an analyses of standard economc approach to savng mples that people wth hgh temporary ncome wll save more to compensate for lower future ncome, wth those wth temporarly lower ncome tend to save less n antcpaton of hgher future ncome. Even though savng rates are hghly volatle wth lfetme ncome, ndvduals wth hgh current ncome, ceters parbus, are expected to be thrfter than those wth low current ncome (Fredman, 1957). The Permanent Income Hypothess follows the logc of the Lfe Cycle Hypothess; that consumpton should not depend on current ncome alone. Unlke the Lfe Cycle Hypothess, whch antcpates that ncome follows a regular pattern over a lfetme, the Permanent Income Hypothess nstead emphaszes that people experence random and temporary fluctuatons n ncome from year to year. Fredman dvdes ncome nto a permanent part, whch s the ncome people expect to persst nto the future, and transtory part, whch s the ncome that people do not expect to persst. More smply, permanent ncome s average ncome and transtory ncome s devatons from ths level of ncome. Fredman argued that consumpton should prmarly depend on permanent ncome because people save and borrow to smooth consumpton n response to transtory changes n ncome. If shocks to ncome are permanent, then all future levels of ncome wll be revsed upwards or downwards by the same amount leadng to a change n consumpton, whch s equal to the change n current ncome. Central s that the consumpton plan does not depend on the transtory components. Fredman argued that ndvduals wth hgh permanent ncome consume the same fracton of permanent ncome as ndvduals wth low permanent ncome. DeJuan et al. (2006) and also Dynan et al. (2000) took up several studes of the Permanent Income Hypothess and conclude that the evdence for the hypothess s mxed at best, wth some studes supportng t and some studes reectng t. A problem wth the theorem, emphaszed by Meghr (2004), s that the loose defnton of permanent ncome makes t dffcult to measure. However, the reason that the Permanent Income Hypothess has endured so much s that, beyond ts smple ntutve appeal, t focuses on nter-temporal optmzaton of consumer behavor, whch s logcal and consstent. The determnants of savngs generally and the specfc effects of government polces on savngs s pvotal force n nvestment and economc growth. The level of savngs depends on the margnal propensty to save of the players of the economy; ths s the proporton of a small change n dsposable ncome that would be saved, nstead of beng spent on consumpton. It s computed by dvdng the change n savngs by the change n dsposable ncome that caused the change. Savngs also depend on the wllngness and ablty of ndvduals and corporate to save. Few studes nvestgated ndvduals savngs usng mcro econometrc approach due to lack of data. Accordng to our knowledge, household savng behavour n Zmbabwe at mcro level has been carred out. From the few studes that have analysed household savng behavour at mcro level, some of the ncluded varables are as follows. 2.3 Income Growth and Income Level Income s also a maor determnant of the level of savngs. Subsstence-consumpton theores suggest that countres wth hgher ncome levels tend to have a hgher savng rate; the aggregate savngs wll ncrease n response to an ncrease n ncome growth and vce-versa when ncome s low consumpton tends to be hgh. 2.4 Demographc Varables Age and geographcal dstrbuton tend to have an mpact on savngs. Consumers are consdered to have negatve 161

4 Internatonal Journal of Busness and Management Vol. 8, No. 10; 2013 savngs when young and wth low ncome, postve savngs durng ther productve years and negatve savngs when they are old and retred (Modglan, 1970). Another demographc varable s the urbanzaton rato, defned as the percentage of the total populaton lvng n urban areas. Ths varable s also expected to have a negatve mpact on savng 2.5 Employment Ths s an mportant determnant of savngs snce the more the people employed the more ther ablty to generate ncome that would be saved f not fully consumed, comng from another angle, f there are more people employed at one household ths means that the ncome of one person s not exhausted n consumpton but can be partly consumed and the remander s postponed It s clear that varous nfluences are suggested but unfortunately these provde ambguous results. Our econometrc specfcaton allows us to test responses of selected ndvdual characterstcs whch were nputted n the logt model as savngs predctors were age, martal status, relgon, educaton, poston n the household, household sze, type of accommodaton, place of accommodaton, employment status, ncome, number of people employed n the household, and expendture per month. In addton, we nvestgated wthn a sample of savers what determned the number of tmes the ndvdual depost money n the bank by use of Posson regresson model. 3. Data and Methodology 3.1 Emprcal Model Savng can be defned as the change n wealth over a certan tme perod or as the dfference between ncome and consumpton (Beverly et al., 2003). It s a complement to consumpton, f ncome and wealth are held constant (Xao and Olson, 1993). In ths research, we used a logt model and a Posson regresson model. For the logt model we generated the savng varable by frst subtractng amount consumed from the household head s dsposable ncome. Income n excess of consumpton was used n the current study to create a dchotomous depended varable; SAVED, coded as 1 f the amount saved s at least US$100 and 0 otherwse. The logt model s used to come up wth the savngs predctng model. The logstc functon s as follows Clearly, 0 f(s) 1 The varable S represents the savngs determnng factors whle f(s) represents the probablty of an event occurrng. The varable S s a measure of the total contrbuton of all the savngs determnng factors used n the model and s known as the logt. The above functon can be transformed nto the followng: S X 0 (1) Where: S = savngs; β 0 = automous savngs; β = vector of regressors parameters; X = a vector of soco-economc and demographc characterstcs and ε = error term. Usng the logt regresson model gven above, the lkelhood of savng s modeled as a functon of the household s ncome and several other covarates prescrbed by lterature to affect one s savng behavour. The Posson regresson model s a count model whch we used to regress the number of tmes one makes deposts. The model n ts complete verson s as follows: NS X 0 (2) NS s number of tmes one makes a depost wthn a month. Other parameters are as defned earler on. 3.2 Posson Estmaton Methodology The response varable NS, the number of tmes one makes a depost wthn a month, s a non-negatve nteger or count whch s explaned by a set of covarates (Xs) beng explanatory varables from equaton 2. Accordng Cameron and Trved (2005), Posson s a pont process for the occurrence of the event of nterest. Ths mpled the Posson dstrbuton modeled the number of tmes one makes depost (NS), wth densty 162

5 Internatonal Journal of Busness and Management Vol. 8, No. 10; 2013 ns e Pr( NS ns) ; ns= 1,2, 3 (3) ns! Where s the ntensty or rate parameter. The dstrbuton s refered as P ( ). Based on Green (2005:121), the two moments whch show equdsperson are E (Y ) andvar (Y ) (4) Posson regresson s one of the robust models for the analyss of dscrete data that are based on the assumpton that the dependent varable NS s dstrbuted as Posson, and ts logarthm s a lnear functon of the ndependent varables (Green 2008a), whch by parametersng the relaton between the mean parameter μ and covarates gves: Where s the expected number of tmes ndvdual household save. 0 s the ntercept; k o x e =1, 2,, n (5) ' s and X ' s are vectors of regresson coeffcents and explanatory varables from equaton (2) of model specfcaton respectvely and k s the number of lnear ndependent covarates n ths case equal to 13. Snce k o x Var( y / x) e, by 4 and 5, the Posson regresson model s heteroskedastc, whch needs maxmum lkelhood estmaton (MLE) to correct, whose log-lkelhood functon accordng to Cameron and Trved, 2005: 667, s as follows: ln L( ) y X exp( X ln y (6)! The Posson MLE, denoted by ˆ p s the soluton to k non-lnear equatons (Green 2008a), correspondng to the frst-order condton for maxmum lkelhood, n 1 ( y exp( X )) X 0 (7) Snce X coeffcets ( s) ncludes a constant term the resdual y exp( X ) sum to zero by (7). The 0 log-lkelhood functon s globally concave; hence solvng these equatons Gauss-Newton or Newton-Raphson teratve algorthm yeld unque parameter estmates, whch are both consstent and asymptotcally effcent. By standard maxmum lkelhood of correctly specfed models, Cameron and Trved (2005) asserts that the estmator ˆ s consstent for of each parameter and asymptotcally normal wth the sample covarance p matrx 1 ˆ ( ) n Var p X X (8) 1 In the case where s of the exponental form (5). The frequency of savng by each household s a dscrete varable, and the dstrbuton savng between dfferent households demonstrates asymmetrcal dstrbuton wth a modal value of 0, supportng the approprateness of modelng savng frequency as a Posson process. On the nterpretaton of Posson regresson coeffcents, suppose s the coeffcent of a 0-1 dummy varable lke 163

6 Internatonal Journal of Busness and Management Vol. 8, No. 10; 2013 employment status or gender, then the exponentaton of ths coeffcent (.e. exp( ) represents the proportonate change n the frequency of savng when the value of the varable changes from 0 to 1, ceters parbus. Smlarly, when s a Posson regresson coeffcent for a contnuous varable, lke households age, then exp( ) represents the proportonate change n the mean number savng tmes when the value of the predctor varable ncreases by one unt, holdng everythng else constant (Wang and Famoye, 1997). The above nterpretatons are based on the varable coeffcent sgn and magntude, but they can be nterpreted n terms of ncdence ratos (Greene 2005). Before we nterpret the coeffcents n terms of ncdence rate ratos, we must address how we can go from nterpretng the Posson regresson coeffcents as a dfference between the logs of expected counts to ncdence rate ratos (IRR). In the nterpretaton above, Posson regresson coeffcents were nterpreted as the dfference between the log of expected counts, where formally, ths can be wrtten as log( X 1) log( X), where β s the regresson coeffcent, μ s the expected count of chldren ever born and the subscrpts represent where the predctor varable (regressors), say X, s evaluated at X and X 1 (mplyng a one unt change n the predctor varable X). Recall that the dfference of two logs s equal to the log of ther quotent, X 1 log( X 1) log( X) log( ), and therefore, we also nterpreted the parameter estmate as X the log of the rato of expected counts: Ths explans the "rato" n ncdence rate ratos. In addton, what we referred to as a count can also be called a rate. By defnton a rate s the number of events per tme (or space), whch our response varable, savng frequency qualfes. Hence, we also nterpreted the Posson regresson coeffcents as the log of the rate rato: Ths explans the "rate" n ncdence rate rato. Fnally, the rate at whch events occur s called the ncdence rate; thus we arrve at beng able to nterpret the coeffcents n terms of ncdence rate ratos from our nterpretaton above. IRR enable us to compare and quantfy the mpact that an explanatory varable, takng nto consderaton ts composton, have on the dependent varable. For example f a varable s bnary lke gender, t tells us how many tmes male household heads saves as compared to ther female counterparts, ceters parbus. Accordng to Cameron and Trved (2005), for any model wth exponental condton mean, of a lnear model E( y/ x) X, the coeffcents βs are readly nterpreted as the effect of a one-unt change n regressors on the condtonal mean. Its dfferentaton yelds: E( y / x) exp( X ) (9) x Where the scalar x denotes the th regressor, measures the relatve change n E( y/ x) nduced by a unt change n x, fx s measured on log-scale, s elastcty. For the purpose of reportng a sngle response value, a good canddate s an estmate of the average response, 1 n E( y / x ) 1 ˆ exp(ˆ X ) snce the Posson x n Regresson Model has an ntercept ncluded t smplfes to ˆ y. From 9, f s twce as large as k, then the effect of changng the th regressor by one unt s twce that of changng the k th regressor by one unt. 3.3 Dagnostc Tests The researcher only resort on one dagnostc test of over-dsperson as propounded by Cameron and Trved (2005). It determnes whether the mean and varance are same, a case known as equ-dsperson. If that s the case, a Posson Regresson Model becomes the most approprate estmaton methodology and any form of devaton from ths condton; a Negatve Bnomal Model becomes the most sutable estmaton methodology. The test goes hand n hand wth maxmum lkelhood and ts P-value and sometmes pseudo R-squared. 164

7 Internatonal Journal of Busness and Management Vol. 8, No. 10; Data Analyss and Results Table 1. Descrptve statstcs Varable Observaton Mean St Dev Mn Medan Max Age Income Expendture Savngs On average the respondents were 39 years old, wth the average ncome beng $ wth the maorty havng ncome of $ The results also revealed that n Zmbabwe the amount of expendture was around $ per month wth the correspondng average savng of $ Amongst the households, the lowest savng level was -$ and the hghest reached $ though the maorty saved $ The hgh negatve rate of savng ndcated the dependency of respondents on borrowngs and socal clubs actvtes. 4.1 Dagnostc Test Dagnostc tests were carred out before the estmatons of model as presented below. () Heteroscedastcty Test We used the Breush-Pagan test to test for the presence of heteroscedastcty. The results were as follows: P The null hypothess of homoscedastcty s that there s constant varance. The results do not reect the null hypothess ndcatng that there s no problem of heteroscedastcty. () Multcollnearty Test There was no problem of multcollnearty, see appendx 1 for detaled results. () Equ-dsperson Tests We tested the null hypothess that there was equ-dsperson of the mean and varance of the model, aganst the alternatve that the data portray under-dsperson or over-dsperson characterstcs. The followng results were obtaned: Lkelhood-rato test of alpha=0: chbar2(01) = 2.8e-03 P-value (chbar2) = Based on the result, we do not reect the null hypothess of equ-dsperson, snce alpha coeffcent of has P-value for the ch-square test of whch were found to be nsgnfcant at 5%. The data does not suffer from under or over dsperson. Hence, havng found such data characterstcs the approprate estmaton procedure, whch wll yeld consstent and asymptotcally effcent parameters, accordng to Cameron and Trved (2005), was the Posson Regresson Model nstead of a Negatve Bnomal Regresson Model. 4.2 Logt Model of Results Number of obs = 295 LR ch2(13) = Prob > ch2 = Log lkelhood = Pseudo R2 = Table 2. Logt estmaton results Varable Coeffcent 0dds-ratos Margnal-effects P-value Constant Age Gender Martal status

8 Internatonal Journal of Busness and Management Vol. 8, No. 10; 2013 Relgon Educaton Level Poston n Household Household Sze Type of Accommodaton Place of Accommodaton Employment Status Number Employed LogIncome Account Type Relgon, household sze accommodaton type, place of accommodaton and ncome creases the household head s probablty of savng. But factors such as gender, age, martal status, educaton level household s status, employment status, number of people employed n the household and bank account type owned reduces the probablty of the household head s probablty of savng. To test for the lfe cycle hypothess we used age as an explanatory varable. The results of the study show that there s a negatve and sgnfcant relatonshp between age and savng. Age was found to reduce the probablty of savng whch s contrary to the paradox of thrft whch asserts that an ndvdual becomes thrfty wth age. Odds-rato fgure for age ndcate that an ndvdual n hgher age group saves tmes less than that of one n a lower age group. A unt ncrease n ones age changes the savng rate %. Income hghly and sgnfcantly ncreases the probablty of savng. A $100 change n ones ncome changes the savng rate by 4.4%. One wth hgher ncome saves approxmately 10 tmes the amount saved by low ncome earner. From the emprcal results, gender, martal status, household sze, place of accommodaton and number employed to our surprse dd not sgnfcantly affect savngs n Zmbabwe at 5% but at 10%. A dscrete change n the household head s poston wll change the savng rate by 1.2%. Current account holders, accordng to ths research, has a 39.62% savng rate lower than that of savngs account holders. 4.3 Posson Regresson Number of obs = 294 LR ch2(15) = Prob > ch2 = Log lkelhood = Pseudo R2 = Table 3. Posson regresson results Varable Coeffcent Irr Z-Score P-value Constant Age Gender Martal Status Relgon Level of Educaton Poston of Household Household Sze Type of Accommodaton Place of accommodaton Employment Status Number employed Account Type Log Income Age Squared GenderLnIncome AgeSavng EducatonIncome

9 Internatonal Journal of Busness and Management Vol. 8, No. 10; 2013 The number of tmes one deposts money n the bank s determned by age, gender, martal status, level of educaton, household sze, type of accommodaton, place of accommodaton, ncome, account type, nteracton of gender and ncome, nteracton of age and savng and educaton ncome nteractons. Relgon, poston of household, employment status, number employed and age squared were not sgnfcant n explanng n explanng ndvduals savngs behavour n Zmbabwe. Based on ncdence rate ratos, t ca be conculuded that, contrary a pror postulaton males savng rate s 6,75% less than that of females.sngle household heads savng rate s 13.43% hgher. an addtonal member n the household ncreases the the savng rate by 7.61% mplyng that household heads mannng large household szes saves more as compered to those responsble for small household sze dependng on them. Household heads who own houses are tmes less lkely to save compare to those who do not have such type of ownershp. Income s the hghest sgnfcant determnant factor of tmes of savng wthn a specfed month. savng tmes have an a hghly elastc response to small change n ncome, that s, household heads wth hgh ncome has a savng rate 49.32% hgher than those wth low ncome.ths s n lne wth the Permanent Income Hypothess and Lfe Cycle Hypothess. Interestngly, there was an nteractve effect between gender and ncome that s they have a combned effect on determnng frequency of savngs. For the same level of ncome male has a 49% hgher savng frequency than female wthn a gven month. agan age and savng have an nteractve effect on tmes of savng, young adult households are usually more actve than older once and the resultas has shown that they save more, confrmng the paradox of thrft, that s, people become more thrfy wth age and so ther frequency of savng wll be hgh. But the combned effect of educaton and ncome on household head s frequency of savng was found to be nsgnfcant. 4.4 Savngs Predctng Model The equaton of the fnal model of sgnfcant varable s: Nˆ S.0.064A 1.27R 0.458ES 0.642PH 0.609E 2.405lnY AT If a bank for example has a clent wth the followng: clent s age=44 years, a Chrstan, employed, husband n the household, have a postgraduate qualfcaton, an ncome amountng to 750 and have a savngs account, the savng score s , that s the chance of ths clent savng s 99%. 5. Conclusons Ths study nvestgated the determnants of savngs n Zmbabwe after the adopton of the multple currency regmes. A mcro econometrc approach whch took nto account the ndvduals heterogenety was adopted. The obectve was to understand whch varables authortes and commercal banks would need to target n order to promote a savngs culture. Savngs were determned by age, educaton level, relgon, poston n household, employment status, ncome and account type. From the emprcal results, gender, martal status, household sze, place of accommodaton and number employed to our surprse dd not sgnfcantly affect savngs n Zmbabwe. Secondly, we nvestgated wthn a sample of savers what determned the number of tmes that an ndvdual depost money n the bank by use of Posson regresson model. The number of tmes one deposts money n the bank s determned by age, gender, martal status, level of educaton, household sze, type of accommodaton, place of accommodaton, ncome, account type, nteracton of gender and ncome, nteracton of age and savng and educaton ncome nteractons. Relgon, poston of household, employment status, number employed and age squared were not sgnfcant n explanng n explanng ndvdual s savngs behavour n Zmbabwe. Lastly, we came up wth a savngs predctng model to be able to dentfy targeted customers based on the ndvdual characterstcs. References Attanaso, O., & Banks, J. (2001). The Assessment: Household Savng- Issues n Theory and Polcy. Oxford Revew of Economc Polcy, 17, Beverly, S. G., McBrde, A. M., & Schrener, M. (2003). A Framework of Asset-Accummulaton Stages and Strateges. Journal of Famly and Economc Issues, 24, Cameron, C., & Trved, P. (2005). Mcroeconometrcs: Methods and Applcatons. Cambrdge: Cambrdge Unversty Press. Deaton, A. (2005). Franco Modglan and the Lfe Cycle Theory of Consumpton. Dscusson Paper, Prnceton Unversty, Deaton, A., & Paxson, C. (1999). Growth, Demographcal Structure and Natonal Savngs n Tawan. Prnceton Unversty, pp

10 Internatonal Journal of Busness and Management Vol. 8, No. 10; 2013 Dynan, K., Sknner, J., & Zeldes, S. (2004). Do the Rch save More? Journal of Poltcal Economy, 112, Fsher, P. J., & Montalto, C. P. (2010). Effect of Savng Motves and Horzon on Savngg Behavours. Journal of Economc Pschology, Fredman, M. (1957). A Theory of the Consumpton Functon. Prnceton: Prnceton Unversty Press. Greene, W. (2008). Econometrc Analyss (6th ed.). Prentce Hall: Englewood Clffs. Greene, W. H. (2005). Functonal Form and Heterogenety n Models for Count Data. Foundatons and trends n economcs, 1(2), Jappell, T., & Pagano, M. (1998). The Determnants of Savng: Lessons from Italy Centre for Studes n Economcs and Fnance. Workng Paper no. 1, pp Mankw, G. (2007). Macroeconomcs. The Unted States: McGraw-Hll. Metn Ozcan, K., et al. (2003). Determnants of Prvate Savngs Behavour n Turkey. Appled Economcs, 35, Mnstry of Fnance, Varous Issues. Modglan, F., & Brumberg, R. (1954). Utlty Analyss and Consumpton Functon: An Interpreteaton of Cross- Sectonal Data. In Kurhara, K. (Ed.), Post-keynsan Economcs. New Brunswrk: Rutgers Unversty Press. Modglan, F. (1970). The Lfecycle Hypothess of Savng and Inter-country Dfferences n the Savng Rato. Oxford Unversty Press. Reserve Bank of Zmbabwe, Varous Issues. Stock, J., & Watson, M. W. (2003). Introducton to Econometrcs. The Unted Stated: Pearson Educaton. Wang, W., & Famoye, F. (1997). Modelng Household Decsons wth Generalzed Posson Regresson. Journal of Populaton Economcs, 10(4), Wlson, S. J. (2000). The Savngs Rate Debate: Does the Dependency Hypothess Hold for Australa and Canada? Xao, J., & Olson, G. I. (1993). Mental Accountng and Savng Behavour. Home Economcs Research Journal,