-mail: publications@imf.org Web: http://0-www-imf-org.library.svsu.edu Price: $18.00 per printed copy International Monetary Fund Washington, D.C . © 2021 International Monetary Fund Title page MALTA SELECTED ISSUES August 31, 2021 Approved By European Department Prepared by Michelle Tejada and Yifei Wang. Contents NOWCASTING THE MALTESE ECONOMY A. Background B. Methodology and Data C. Model Performance D. Applying the Nowcasting Models and Conclusions References FIGURE 1. Backtesting Results of Selected Methods 2. Real Time Nowcasts TABLES
observation of GDP data). Our variable of interest here is the year-over-year growth of GDP (in percent). Each “as-if” vintage is named by a quarter, meaning that this vintage contains data as if they are available or released by that quarter, except for GDP. For example, 2020:Q3 vintage contains all variables by the end of September 2020, except for the GDP in 2020:Q3 (GDP is up to 2020:Q2). Re-estimate each model with the “as-if” vintages in expanding windows, and generate 1-step-ahead nowcasts at each quarter to get the time series of backtesting results for each
exercises show how the decline in historical asset volatilities affects the number of VaR exceptions. Using the HS method and constant proportions of assets in the broad portfolio, there is a clustering of exceptions during the turbulent 1997–98 episode ( Figure 2.2 ) and a paucity of violations during the recent calmer period ( Figure 2.3 ). Figure 2.2. Backtesting Results: Broad Portfolio, October 1997 to October 1998 Sources: Bloomberg L.P.; and IMF staff estimates. Note: HS VaR = historical simulation of value-at-risk. Yellow squares indicate VaR
.91 GPReg 10.73 4.02 3. An Integrated Tool Using the methodologies described in this paper, we developped an integrated tool that automatically (i) collect and treat the data set; (ii) applies a suit of DMF and ML models to the dataset to generate backtest results (graphs and quantitative indicators), and (iii) nowcast GDP growth for the current quarter for each method and aggregating all the methods. The tool can be applied to any country, and is automated once variables to form the dataset have been selected. The tool is developed in Matlab
and Elastic Net) and two types of tree-based ensemble the models (RF and RUSBoost). The signal extraction method is superior in terms of both out-of-sample AUC and sum of errors. No single model consistently performs the best in predicting EMP events . In backtesting, the signal extraction model performs the best (in terms of sum of errors) for AEs, RUSBoost performs the best for EMs, and RF model performs the best for LICs. Implied by these backtesting results, a signal extraction model is applied to predict EMP events in AEs and two RF models estimated within
amendment is based on a proposal by the Basle Committee on Banking Supervision published in January 1996, 24 and covers only large banks with significant trading activities. The proposal provides additional guidance to banking institutions on how backtesting results will be directly linked to an institution’s potential capital charge. Backtesting, which is comparing VAR amounts generated by internal models (institutions must use VAR amounts generated for internal risk-measurement purposes, not the daily VAR generated for supervisory capital purposes) against actual
the standardized approach and the Internal Models Approach) and BA 325 (daily return that covers selected risk information, including the standardised approach and the Internal Models Approach) are periodically analyzed and compared with bank’s trading balance sheets, limit structures and financial performance. For internal model users, backtesting results are regularly provided to the BSD and in case these results indicate poor model specification, the models capital requirement multiplier will be adjusted. Assessment Compliant. Comments