Working Paper Series: Special Edition of 2016 to 2018 Interns

bank will hold these items on their books as collateral via bill of sale. If these items were originally purchased from retained earnings, it means that if the firm decides to borrow in the future it would have collateral that could be used to secure loans and advances.

4 Econometric Methodology: 4.1 Model Specification

In specifying a model, one seeks to develop a construct that is able to capture the dynamics of a given process. Ideally, the model should not be one that is exhaustively over-specified nor under- specified but adequately specified in the sense that it includes all the relevant variables in line with prevailing literature, economic theory and logical reasoning in relation to the area of inquiry. Exactly specifying a model is usually difficult to achieve in practice due to noise, errors in the data and missing data points. In the spirit of general to specific modelling, if one must mis-specify the model, it is usually preferable to over-specify the model and refine it by pruning variables, as opposed to under-specifying the model. The preferred model of choice for answering the research questions is the logistic regression model. As Nu Minh Le (2012) posits, most conventional studies and traditional research on SME access to bank credit either uses the probit or logit models. It should be noted that it is possible but not a necessary condition to derive binary choice models from an underlying latent model framework (Verbeek, 2004). The logit model follows what is known as the cumulative density function (cdf). The generic setup of the logistic model is specified in equation 1. Pr( = 1| ) = exp( ′ ) 1 + exp( ′ ) = ᴧ ( ′ ) ……… (1) For the analysis using the full sample of 1966 firms and a sub-sample of the ECCU of 772 firms, two models were estimated. These are specified as follows: , � + The outcome variable bank credit is dichotomous in nature identifying whether the ℎ firm located in country j has a loan or not. Firm Specific is vector representing a set of variables that include key policy variables as well as other that include the characteristics specific to the firm (technical, foreign, larger, size, legal, education). Bank Assessment is a vector representing a set of credit assessment characteristics (innovation, overdraft, collateral, development, marketing, 1. Pr( ) = � ,

20