RDP 2024-08: Modelling Reserve Demand with Deposits and the Cost of Collateral 3. Literature Review

The literature identifies three unique features about the demand curve for reserves and several techniques for estimation consistent with these features.

First, it is highly nonlinear. The reserve demand curve is frequently drawn as kinked or roughly hyperbolic around a certain high level of reserves – known as the ‘saturation’ point (Frost 1971; Reis 2016). Beyond the saturation point banks are no longer willing to pay above the ES rate to obtain more reserves. The literature assumes various functional forms which differ in the degree to which market rates rise as reserves decline from their saturation point, including: log, hyperbolic, logistic, double exponential and arctangent (Nicolae 2020; Veyrune, della Valle and Guo 2018; Smith 2021; Bräuning 2018; Chen, Kourentzes and Veyrune 2023). In support of the log functional form, Lopez-Salido and Vissing-Jørgensen (2024) derive a microfoundation of the convenience yield of reserves in which reserves enter in logs. The logistic functional form is unique among the related literature as it assumes an upper asymptote as reserves decline toward zero. Veyrune et al (2018) impose the upper asymptote at the lending facility rate due to a lack of observed data close to this rate – a strong identifying assumption which presupposes an absence of stigma at the lending facility.

Second, banks' demand curve for reserves has been shown to shift over time. To document a shift, one strand of the literature estimates the demand curve separately for distinct time periods such as during central bank balance sheet expansion and unwind (Aberg et al 2021; Smith 2021). Another strand uses high-frequency data and rolling time periods to estimate the slope of the demand curve over time (Afonso et al 2022; Smith and Valcaral 2023). Both approaches are useful in documenting shifts in the demand curve and in monitoring how steep the slope of the demand curve is in real time (Afonso et al 2024). They do not, however, explain why the demand curve has shifted. Using US data, Lopez-Salido and Vissing-Jørgenson (2024) confirm empirically that liquid deposits shift banks' demand curve in an ample reserves system. They conclude that banks are willing to pay for additional reserves if they reduce transaction costs associated with managing a higher volume of payments between depositors, or in other words additional reserves generate a ‘convenience yield’. The finding that liquid deposits shift banks' demand curve holds when replicated using UK and euro area data but not Canadian (Meinecke 2023; Vissing-Jørgensen 2023; Bulusu et al 2023). Using US data, Acharya et al (2023) find that credit lines are statistically significant, in addition to liquid deposits, when explaining a shift in reserve demand. The mechanism for credit lines affecting demand is conceptually similar to liquid deposits but credit lines emphasise the importance of banks' holding precautionary buffers against claims on bank liquidity.

Third, estimating demand is likely subject to endogeneity – regressing the price of reserves on their quantum only reveals the demand curve if all changes in the quantity of reserves represent shocks to supply. Said differently, estimating demand requires the supply of reserves to be exogenous. In a demand-driven system, the quantity of reserves is not exogenous to demand as banks changing their borrowing from OMO influences the supply of reserves. I address endogeneity using an IV approach tailored to the Australian context in Section 6.