SearcharxivSearch

arXiv subjects

Xuan Mei

Publications and source records attributed to Xuan Mei.

4 recordsLinked to original sources

Attributing Differences Between Forecast Runs to Input Changes, With Applications to CCAR and CECL Exercises

Forecasting systems used in the Comprehensive Capital Analysis and Review (CCAR) and Current Expected Credit Losses (CECL) processes combine portfolio data, macroeconomic scenarios, model specifications, business assump- tions, and management adjustments. When the forecast changes from one run to the next, practitioners need an attribu- tion that reconciles to the total change without depending on an arbitrary sequence of input replacements. This paper formulates forecast-gap attribution as a cooperative game and examines several approaches: the exact Shapley value, hierarchical or nested Shapley values, Integrated Gradients, Gradient SHAP, Permutation SHAP, and Kernel SHAP. We compare their allocation rules, computational costs, implementation requirements, and limitations in production forecasting systems. The analysis provides a practical framework for choosing an attribution method according to the number and type of inputs, the feasibility of hybrid forecast runs, and the need for interpretability, reproducibility, and governance.

q-fin.RM

Attributing Forecast Gaps to Component Models in Complex Model Suites

Complex model suites composed of multiple interacting component models are widely used in financial forecasting and risk management. In model performance testing, including in-sample backtesting (BT) and out-of-sample ongoing performance monitoring (OPM), a material gap between a model-suite forecast and the realized outcome must often be attributed to individual component models for development, validation, and regulatory review. This paper studies this gap-attribution problem in the expected loss framework, where exposure at default (EAD), prepayment or single monthly mortality (SMM), probability of default (PD), and loss given default (LGD) interact multiplicatively and are aggregated across loans and projection periods. We first formalize standard walk analysis and show why its attribution is generally order dependent. We then adapt two order-independent attribution frameworks: an augmented Logarithmic Mean Divisia Index (LMDI) approach tailored to the expected-loss structure, and a more general Shapley value approach based on averaging marginal contributions over all component orderings. We derive both elementwise and vectorized formulas to support efficient implementation, with the additional computation time for gap attribution typically limited to a few seconds in practical portfolio-scale examples. Finally, we discuss the connections among walk analysis, LMDI, and Shapley attribution, and show how the attribution framework extends to model suites with an additional Monte Carlo simulation layer.

q-fin.RM

Using CPI in Loss Given Default Forecasting Models for Commercial Real Estate Portfolio

Forecasting the loss given default (LGD) for defaulted Commercial Real Estate (CRE) loans poses a significant challenge due to the extended resolution and workout time associated with such defaults, particularly in CCAR and CECL framework where the utilization of post-default information, including macroeconomic variables (MEVs) such as unemployment (UER) and various rates, is restricted. The current environment of persistent inflation and resultant elevated rates further compounds the uncertainty surrounding predictive LGD models. In this paper, we leverage both internal and public data sources, including observations from the COVID-19 period, to present a list of evidence indicating that the growth rates of the Consumer Price, such as Year-over-Year (YoY) growth and logarithmic growth, are good leading indicators for various CRE related rates and indices. These include the Federal Funds Effective Rate and CRE market sales price indices in key locations such as Los Angeles, New York, and nationwide, encompassing both apartment and office segments. Furthermore, with CRE LGD data we demonstrate how incorporating CPI at the time of default can improve the accuracy of predicting CRE workout LGD. This is particularly helpful in addressing the common issue of early downturn underestimation encountered in CRE LGD models.

q-fin.RM

An Exploration to the Correlation Structure and Clustering of Macroeconomic Variables

As a quantitative characterization of the complicated economy, Macroeconomic Variables (MEVs), including GDP, inflation, unemployment, income, spending, interest rate, etc., are playing a crucial role in banks' portfolio management and stress testing exercise. In recent years, especially during the COVID-19 period and the current high inflation environment, people are frequently talking about the changing "correlation structure" of MEVs. In this paper, we use a principal component based algorithm to perform unsupervised clustering on MEVs so we can quantify and better understand MEVs' correlation structure in any given period. We also demonstrate how this method can be used to visualize historical MEVs pattern changes between 2000 and 2022. Further, we use this method to compare different hypothetical and/or historical macroeconomic scenarios and present our key findings. One of these interesting observations is that, for a list of 132 transformations derived from 44 targeted MEVs that cover 5 different aspects of the U.S. economy (which takes as a subset the 10+ key MEVs published by FRB), compared to benign years where there are typically 20-25 clusters, during the great financial crisis (GFC), i.e., 2007-2010, they exhibited a more synchronized and less diversified pattern of movement, forming roughly 15 clusters. We also see this contrast in the hypothetical CCAR2023 FRB scenarios where the Severely Adverse scenario has 15 clusters and the Baseline scenario has 21 clusters. We provide our interpretation to this observation and hope this research can inspire and benefit researchers from different domains all over the world.

q-fin.RM