Introduction

We provide cloud-based risk analytics for corporate bonds with a focus on the Chinese market. We employ a Big Data/Artificial Intelligence framework with our own proprietary models and machine-learning algorithms. Our approach combines several state-of-the-art modeling techniques including financial statement analysis, Contingent Claims analysis of default risk, balance sheet misreporting risk, and natural language processing of news and Internet comments.

1. Bond data search and queries 
2. Credit risk modeling and ratings 
3. Yield decomposition into risk compensation and liquidity premium
3. Matrix pricing of illiquid bonds
4. Market monitoring
5. Portfolio analysis including calculation of economic capital 

Model-Based Bond Selection Outperforms Benchmark and Major Global Indices

Creating market liquidity is the value proposition

Today a few hundred bonds are traded…

…out of more than 5,000 corporate bonds

The greatest opportunity since the USA in the 1980s

China has a nearly $10 trillion and gorwing corporate bond market. Most bonds don’t trade, and price discovery is difficult. Lack of information, poor liquidity, and negative credit surprises discourage foreign investors. China’s corporate market resembles the American mortgage market in the early 1980s. The advent of mortgage securitization and widely-accepted valuation models led to a large and liquid market in mortgage-backed securities. Increased market efficiency drew new classes of domestic and foreign investors to the mortgage-backed securities market and increased the efficiency of US capital markets. Investment banks who mastered securitization and modeling earned enormous rewards.

Harvesting value from enhanced liquidity and investor participation

Inadequate risk assessment makes the great majority of the CNY corporate bond market illiquid.
Foreign investors are reluctant to commit to a market with poor transparency and many negative surprises.
Greifenberg’s Machine-Learning Model can create transparency, provide accurate valuations for CNY bonds that now are illiquid, and expand the investor base for higher-yielding CNY bonds.
GFC’s Machine-Learning Model approach combines three proven technologies (Financial Statement analysis, Contingent Claims analysis, and Big Data analysis of financial statement reliability) with a key innovation (National Language Processing of market sentiment). Machine-Learning then integrates these four modeling approaches.
Greifenberg can create an industry standard for transparent and accurate risk assessment.

Merton's Contingent Claim (Option) Approach in Assessing Credit Risk

Robert Merton’s Key Insight:
The holder of a default-prone bond in effect sells a put on the asset value of the bond issuer. If the issuer defaults; it “puts” the assets of the company to the bondholder. This usually occurs when the issuer’s equity price approaches zero. Net, the bond’s value depends on the value of the firm’s assets (the equity price) relative to debt and the volatility of the asset value (approximated by equity volatility). All of these can be observed.

We translate observed balance sheet leverage, equity price and equity volatility into distance to default. This captures some but not all of the pricing dynamics of corporate debt.

Therefore, assessing the credit risk amounts to computing this put option value:

Value of risky debt = Value of risk-free debt – Put Value on the firm’s assets

China vs. US Corporate Bond Trading Volume

Greifenberg Meta-Model: General Framework

Analysis

Pricing Analysis
Matrix Pricing Approach

Find the nearest liquid bonds as reference.
Assign weights among the bonds based on the calculated distances.
Predict yield and transaction-related statics using weights and feature values.
Accurately simulate the target bonds’ yield-implied rating.

Out-of-sample back-testing confirms predictive power

AUC = Area Under Curve

Case Study:
China Fortune Land

In March 2021, China Fortune Land Development, a Chinese property developer defaulted on a $530 million bond in the latest test for the country’s debt-laden real estate sector. While the bond caried AAA rating to the end, our model indicated otherwise . As shown in the chart below, the default risk, calculated by both the Merton Model and Greifenberg proprietary predictive models, has been in an unequivocally upward trend (towards default) in the last few semi-annual periods:

  • The red line in the top panel is the Probability of Default (PD) derived from the GFC Merton model, which rose sharply starting in June 2017, indicating an almost certain default (82%) starting in December 2018, suggesting a “CCC” rating, rather than “AAA”, assigned by CCXI, a large rating agency.
  • A similar conclusion can be reached by using Greifenberg’s first version predictive model using just financial statement data, which shows default risk measured by PD (on a different scale) rising steadily, as is shown by the chart in the lower right corner.

Risk of inaccurate reporting on financial statements rose sharply

Natural Language Processing (NLP) detected a surge of negative commentary two months before the actual default event.

Matrix pricing analysis showed a significant deviation between the bond’s traded price and matrix price by Greifenberg model, highly likely indicating an overvalued bond.  


Website Design  - Search Portal

Bond Search Portal and Sorting Feature:

– Search by bond ticker and issuer name
– Criteria Search by Credit Score, Liquidity (volume), maturity, coupon rate, rating, yield, etc.
– Retrieve the bond analytical detail page by double clicking the bonds’ names
– Intuitive display of bond features and risk return profile
– Early warning of stress
– Watch list

Summary

China’s corporate bond market provides the world’s largest offering of marginal yield to global investors, with an attractive risk/return profile.  Analytical tools are undeveloped for this major market. Chinese rating agencies have been unreliable. Greifenberg offers a unique analytic suite for risk analysis of Chinese corporate bonds. GFC’s Machine-Learning model framework incorporates inputs from four credit risk models using several inputs and methodologies and subjects them to machine learning algorithms.  Our proprietary approach allows us to capture all relevant information efficiently and provide the most reliable credit assessment now available.