art = wandb.Artifact("my-object-detector", type="model")
art.add_file("saved_model_weights.pt")
wandb.log_artifact(art)
CCAR Example
Created on August 1|Last edited on August 8
Comment
Executive Summary
CCAR (Comprehensive Capital Analysis and Review) is a regulatory framework governed by the Federal Reserve to assess, regulate, and supervise large US banks that are too big to fail.
The three main principles for CCAR, born out of the 2008 global financial crisis, are:
- To assess whether banks possess adequate capital for their needs, both now and in the future
- To review the capital structure & stability to various stress-test scenarios.
- To assess planned capital distribution, including shares or dividends, for viability with minimum capital requirements.
The global economic crisis/subprime crisis/recession of 2008 resulted in the economic collapse of some of the largest banks in the US. CCAR is a kind of stress testing steered by the Federal Reserve board. Supervisory stress tests are an integral part of prudential regulations across the number of banks’ jurisdiction. The CCAR report provides a view on the industry facing economic risks, cyber risk, and emerging risks from financial technology companies as a whole and shows a comparison among different institutions.
The focus of this report is tracking of modeling exercises associated with bullet point 2 above: To review the capital structure & stability to various stress-test scenarios. We may consider developing models that link economic indicators to the balance sheet to stress capital ratios in a an effort to understand the banks stability.
Data
For this exercise, two Macroeconomic indictors, Real GDP Growth and Market Volatility Index, were used to generate a fake dataset of loan balance by year to demonstrate how W&B can be used for Regulatory exercises like CCAR and CECL.
is to model loan to loss which is then used to stress the balance sheet in order to understand how an FI's capital ratios make react to shocks in the economy. These shocks, or scenarios, are provided by the OCC on a yearly basis and banks are meant to forecast their balance sheets off of these ratios in an effort to understand any risk to the bank.
We'll pursue stress testing here via a loan to loss model, which would then be used to stress the balance see in order to understand how an FI's captial ratios may react to economic shocks. The model highlighted in this report would correspond to forecasting balance of performaning loans for a particular line of business -> this would balance would be composed of balance for new and existing loans. Later, it may flow through a default model, then a loss given default model. This way scenarios can impact each of the three components which generally impact the bank's balance sheet.
Stage 1: Balance Forecasts
The target in this problem could take many forms, but for our purposes, just suppose that the target is the total balance of active HELOCs. We would typically rely on theory and practical significance to pick the best economic indicators that should best explain our target variable. Moreover, modeling for CCAR and CECL typically rely on methodologies that permit statistical inference. It is OK to use more modern ML methodologies in pursuit of the final models, but at the end of the day, relying entirely on data mining is not favorable.
With that being said, data mining is unavoidable given scarcity of resources and using it to help get started in the right direction is permissible.
This set of panels contains runs from a private project, which cannot be shown in this report
Model Methodology
As mentioned, we are assuming that our target variable is a function of VIX and Real GDP Growth, plus some Autoregressive error structure, or more explicitly
with
with . In order to estimate this model, we will pursue a Generalized Least Squares setup with an AR(1) error Structure. In order to estimate the AR structure, we'll use Yule Walker Method.
Below is the output from the model which was creating in python as well as some plots detailing the path of the structural mean of the target over time under all of the provided CCAR scenarios. On the plot, we have also provided an analysis of the sensitivity of the forecast, under the severe scenario, when we use T-Copula to understand the 5th and 95th percentile of the distribution of the coefficients for the structural mean part of the equation.
This set of panels contains runs from a private project, which cannot be shown in this report
Supervisory Scenarios
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.
Run set
0
gls-model
Artifact overview
Type
model
Created At
August 1st, 2022
Description
Versions
Version
Aliases
Logged By
Tags
Created
TTL Remaining
# of Consuming Runs
Size
No rows found
Loading...
Add versions to this Registered Model
Collect the best candidate models and use aliases to track a model on its journey to production.
Quickstart
- On the top right of this page, click to create a Registered Model.
In your script, log a model as an artifact version.
Code Snippet- From the Artifact page, link the artifact version to the registry.
Try it yourself
Watch this 1-minute video and follow along with this notebook →
Add a comment