Bayesian VAR model forecasts unquoted shares in Italian accounts
Banca d'Italia researcher Michela Eugenia Pasetto has developed a new Bayesian vector autoregression model to forecast unquoted shares of Italian non-financial corporations. The method improves quarterly financial accounts by outperforming five alternative econometric approaches.
Forecasting the unquoted corporate sector
Valuing unquoted shares on the liability side of non-financial corporations is crucial for understanding corporate financing choices.
Because balance sheet data are released annually with a 15 to 18-month delay, quarterly financial accounts rely heavily on estimation techniques.
To address this, Banca d'Italia researcher Michela Eugenia Pasetto proposes a hierarchical Bayesian vector autoregression model based on Giannone, Lenza, and Primiceri.
The approach forecasts both outstanding amounts and transactions in unquoted equity on a quarterly basis, utilizing real and financial aggregates from Cerved covering data from 1995 to 2023 across nearly 908,000 unquoted Italian companies.
Outperforming five alternative models
The estimation process begins with temporal disaggregation of annual balance sheets using the Chow-Lin method with quarterly loan liabilities as the high-frequency indicator.
Priced unquoted firms, representing roughly 70 percent of total unquoted own funds, are subsequently assigned market valuations based on comparable listed companies.
When compared against five alternative econometric approaches, the Bayesian vector autoregression model demonstrates superior out-of-sample forecasting performance for point forecasts and variability.