After discussing various factors related to bank lending, we are now ready to specify variables for the empirical models. Following loan pricing literature(Bharath, et al, 2010 Lin et al, 2011; among others), loan rate can be measured by the spread over a benchmark interest rate. The natural choice of the benchmark rate in china is the benchmark lending rate set by PBC, which is also widely used in loan pricing in practice. The empirical model can be written as Spread t =Bo+B, MP_+B, Mr+B, FCi+B.LCit-l+Bs ADi+a+ur( 8) Where Spread t is the price spread over the corresponding benchmark lending rate for firm i at time t. All explanatory variables are lagged one quarter in order to avoid endogeneity caused by simultaneousity. MP- indicates monetary policy instruments such as benchmark deposit rate and rrr in the period t-l. MRit- represents the free market interest rate which is the market interest rate for bank lending. FCit-denotes various firms characteristics such as the firm's industry, firms ownership, and various financial characteristics such as return of equity, total asset, total employment, liquid asset, debt-to-asset ratio, equity-to-debt ratio and profit margin. LC,- includes features of a specific loan such as loan maturity, the type of bank and collateral. AD- denotes macroeconomic variables including fixed-asset investment growth, money supply and market liquidity. a represents unobservable firm characteristics such as reputation of the firm, relationship between firm and bank etc. u t is the idiosyncratic shock which is not correlated with any explanatory variables Similarly, we can write down an empirical model for loan quantity as follows L=Φ。+Φ1MP1+Φ2MRa-1+中3FC+中LC+中ADn++En(9 Where L is the loan quantity for firm i at time t. u; represents some unobservable fixed effect impacting the loan size and e is the idiosyncratic shock. Other variables are the same as Equation( 8). Since the dataset is a panel dataset, it is good to remove observable time-constant features such as the firms industry and ownership. Moreover, the panel also enables us to eliminate unobservable time-constant features such as the firms' reputation and the relationship between banks and firms. Empirically, the easiest way to remove these fixed effects is to take the first differencing of Equation(8 )as follows △ Spread=B1△MP1+B2△MR1+B3△FC-1+B4△LCi-1+B5△ADn-1+△an(10) In practice, prices in loan contracts in China are as a certain amount of interest rate For different maturities. PbC has set different be rates. The spread of a loan is calculated based on its corresponding benchmark lending rate11 After discussing various factors related to bank lending, we are now ready to specify variables for the empirical models. Following loan pricing literature (Bharath, et al, 2010; Lin et al, 2011; among others), loan rate can be measured by the spread over a benchmark interest rate. The natural choice of the benchmark rate in China is the benchmark lending rate set by PBC, which is also widely used in loan pricing in practice.8 The empirical model can be written as it it it it it it i it Spread = Β + Β MP + Β MR + Β FC + Β LC + Β AD + + u 0 1 −1 2 −1 3 −1 4 −1 5 −1 α (8) Where it Spread is the price spread over the corresponding benchmark lending rate for firm i at time t 9 . All explanatory variables are lagged one quarter in order to avoid endogeneity caused by simultaneousity. MPit−1 indicates monetary policy instruments such as benchmark deposit rate and RRR in the period t −1 . MRit−1 represents the free market interest rate which is the market interest rate for bank lending. FCit−1 denotes various firm’s characteristics such as the firm’s industry, firm’s ownership, and various financial characteristics such as return of equity, total asset, total employment, liquid asset, debt-to-asset ratio, equity-to-debt ratio and profit margin. LCit−1 includes features of a specific loan such as loan maturity, the type of bank and collateral. ADit−1 denotes macroeconomic variables including fixed-asset investment growth, money supply and market liquidity. αi represents unobservable firm characteristics such as reputation of the firm, relationship between firm and bank etc. it u is the idiosyncratic shock which is not correlated with any explanatory variables . Similarly, we can write down an empirical model for loan quantity as follows: Lit MPit MRit FCit LCit ADit i it = Φ + Φ + Φ + Φ + Φ + Φ + µ + ε 0 1 −1 2 −1 3 −1 4 −1 5 −1 (9) Where Lit is the loan quantity for firm i at time t . µi represents some unobservable fixed effect impacting the loan size and it ε is the idiosyncratic shock. Other variables are the same as Equation (8). Since the dataset is a panel dataset, it is good to remove observable time-constant features such as the firm’s industry and ownership. Moreover, the panel also enables us to eliminate unobservable time-constant features such as the firms’ reputation and the relationship between banks and firms. Empirically, the easiest way to remove these fixed effects is to take the first differencing of Equation (8) as follows: it it it it it it it ∆Spread = Β ∆MP + Β ∆MR + Β ∆FC + Β ∆LC + Β ∆AD + ∆u 1 −1 2 −1 3 −1 4 −1 5 −1 (10) 8 In practice, prices in loan contracts in China are typically written as a certain amount of interest rate spreads over the corresponding benchmark lending rate set by PBC. 9 For different maturities, PBC has set different benchmark lending rates. The spread of a loan is calculated based on its corresponding benchmark lending rate