research Market Intelligence /marketintelligence/en/news-insights/research/pd-sentiment-china-corporate-10 content esgSubNav
In This List
Research

PD Sentiment China Corporate 1.0

Case Study

An International Logistics Company Streamlines Transfer Pricing for Its Extensive Network of Subsidiaries

Podcast

Next in Tech | Episode 168: AI Data Strategies

Podcast

Street Talk | Episode 126: Hunting the bears, making the bull case for CRE

Blog

Investment Banking Essentials May 15


PD Sentiment China Corporate 1.0

Highlights

Leveraging fundamental and sentiment data to assess the probability of default of companies in the Chinese domestic market

S&P Global Ratings does not contribute to or participate in the creation of credit scores generated by S&P Global Market Intelligence. Lowercase nomenclature is used to differentiate S&P Global Market Intelligence credit model scores from the credit ratings issued by S&P Global Ratings.

Overview

Thanks to advances in natural language processing (NLP) in recent years, sentiment data has been widely explored to gain information advantage in financial market. Sentiment analysis is often used to uncover signals embedded in a variety of unstructured text corpus, e.g., social media news and company announcements. As a useful complement to the traditional financial statement data, sentiment data may provide new insights on a company’s financial health in a timely manner. At S&P Global Market Intelligence, we have developed PD Sentiment China model (PDS China 1.0), where sentiment data are used as novel factors to gauge the credit risk for a corporate obligor. The model generates 1-year probability of default (PD) for public and private companies in the China market, aiming to provide early warning signals at a short-term horizon prior to credit events.

More specifically, PDS China 1.0 is built on a logistic regression framework using both company sentiment scores and PD estimates from the PD Fundamentals Model 2.0 (PDFN 2.0) China Model. [1] The final PDS model PD is adjusted to reflect the “real” or target default rate that is in line with the benchmark, which is based on the average observed default rate for companies in the Chinese domestic market. The model can be used by risk managers at financial institutions, corporations, and asset management firms to:

  • Capture alternative credit risk signals from large-sized and unstructured text datasets in an automated way.
  • Provide timely early warnings of default events on a daily basis, especially for periods between financial statement filings, which are typically updated quarterly or annually.

Model Highlight

The model is trained on the bond issuers sample in the China local market with default flags sourced from the S&P Global Market Intelligence’s China Credit Analytics Platform (CCAP). The fullowing factors are considered:

1) Baseline PDs (PDFN 2.0) that reflect the long-term financial health of companies, which are captured by such financial ratios as liquidity, leverage, profitability, efficiency and size factor, and industry risk factor.

2) Sentiment information that captures the short-term signal for a company’s financial health and its time-varying fluctuations. We include two types of data:

  • Sentiment scores, which represent the overall sentiment condition of companies embedded on the text corpus from a variety of selected media channels.
  • News tags, which capture occurrence of selected company events that are closely linked to credit risk, such as liquidity issues, debt warning, financial fraud, slump in revenue, abnormal operations, legal disputes, and other relevant events. [2] A negative tag ratio is calculated to measure the relative frequency of these events.

The PDS China 1.0 consists of three sub-models, each generating an output:

  • Based on sentiment score only
  • Based on PDFN PD and sentiment score
  • Based on PDFN PD, sentiment score, and selected tag ratios

Model Features

Entity Coverage

The model applies to both publicly-traded and privately-owned companies in the China corporate sector (see Appendix for details on coverage) as defined by Primary Industry Classifications (PICs). The model is applicable for companies with necessary sentiment data. [3]

Model Outputs

The model’s primary outputs are one-year PDs and the corresponding PD-mapped credit scores. The mapping of PD to credit score is based on S&P Global Market Intelligence's global rated universe and spans a period of more than 35 years (starting from 1981).

Model Inputs

The main inputs for PDS China 1.0 are:

  • PDFN PD: estimated by PDFN 2.0, a widely applicable statistical model, mainly based on company financials spanning two risk dimensions: financial risk and business risk.
  • Sentiment score: calculated by considering all daily news related to target companies, with different weights applied depending on importance and uniqueness of each news. It is normalized to lie between -1 (very negative) and 1 (very positive).
  • Selected tag ratios: calculated to measure the relative frequency of selected negative company events.

The final model results are generated based on exponentially weighted moving averages of sentiment scores and tag ratios.

PD Calibration

The absulute PD values from a statistical credit risk model are usually influenced by the observed default rates of the training sample. A model with good discriminatory power could also generate PD values that are either too high or too low, when compared to historical observed default rates. However, it is difficult to obtain a training dataset that perfectly represents the real world’s default frequencies, due to data availability. This may skew the model outputs to be either more aggressive or conservative, making the final adjustment of the PD level necessary. In PDS China 1.0, we refer to the European Banking Authority Risk Dashboard to calibrate the model outputs.[4]

Early Warning Signals

We tested how sentiment scores fluctuate at different time points prior to default for historical defaults in the China onshore bond market, and compared the trends with those for non-defaulted bond issuers. Figure 1 shows the one standard deviation band of company sentiment scores for each month prior to default. On average, sentiment scores start to deteriorate twelve months prior to default. They tend to move into the negative zone as the time gets close to default. Further, the closer to the default, the steeper the slope. The drop becomes more dramatic at two months prior to default.

Figure 1. Trend of sentiment scores prior to default events

Source: S&P Global Market Intelligence. Data as of September 2021. For illustrative purposes only.

The selected tag ratio fullows the similar trend. Figure 2 shows that on average, the selected negative tag ratio increases from 0.2 at twelve months prior to default to 0.45 at one month prior to default. This deterioration becomes more striking at four months prior to default.

Figure 2. Trend of selected negative tag ratio prior to default events

Source: S&P Global Market Intelligence. Data as of September 2021. For illustrative purposes only.

Data Statistics on Training Sample

The model is trained on historical data of 4,842 bond issuers in China, including 1,648 public issuers and 3,194 private issuers. The sentiment data, including sentiment scores and news tags, spans the period from 2015 to 2021. The average number of daily sentiment scores per year is 82 for companies in the sample and 50% of companies have more than 41 daily sentiment scores per year.

Figure 3 shows the distribution of sentiment scores for the overall sample. The majority of sentiment score are within -0.5 and 0.5.

Figure 3. Distribution of sentiment scores

Source: S&P Global Market Intelligence. Data as of September 2021. For illustrative purposes only.

Model Performance

The discriminatory power of a PD model can be measured via the Receiver Operating Characteristic (ROC) that shows the model’s ability to distinguish obligors by assigning higher (lower) PD values to companies that will likely (not) default in a specific time horizon.

The ROC performances are tested on the bond issuers sample for the three sub-models and they are:

  • Sentiment score only model: 78.9%
  • PDFN + Sentiment score model: 85.1%
  • PDFN + Sentiment score & Tag ratio model: 87.2%

Using sentiment scores and tag ratios enhances the ROC performance by a large margin (87.2% vs. 77.3%), when compared with the base case (PDFN 2.0).

Case Study on an Anonymized Company

Company X is one of the largest real estate companies in China and had been actively expanding its business throughout the country. As of September 2021, the company failed to pay debt that was due.  

Figure 4: Case Study

Source: S&P Global Market Intelligence. Data as of April 2022. For illustrative purposes only.

Figure 4 illustrates the evulution of PD estimates generated by the PDS China model for the period 2020-2021. Sentiment scores have been continuously declining since November 2020 and dropped sharply during July-August 2021. The PDS model PD has been climbing up and jumped to more than 8% in late July. The downward trend of sentiment scores, along with the increasing model PD, provides strong early warning signals for the credit event in September 2021.

Conclusion

Timely and robust credit risk assessments for counterparties in China are challenging due to the low-quality and lag of traditionally used financial statement data, especially for private companies. S&P Global Market Intelligence’s PDS China 1.0 model, utilizing company sentiment scores in addition to company financials, offers an automated and scalable sulution for gauging the short-term credit risk of corporates in the China domestic market. The model can provide strong early warning signals within one year prior to credit events.

APPENDIX

A. PD Sentiment China: Supported Industries (as of April 2022)

Industry

Code

Industry Name

GICS

Description

1

Aerospace & Defense

20101010

Aerospace & Defense

2

Airlines

20302010

Airlines

3

Automotive

25101010

Auto Parts & Equipment

 

 

25101020

Tires & Rubber

 

 

25102010

Automobile Manufacturers

 

 

25102020

Motorcycle Manufacturers

4

Energy

10101010

Oil & Gas Drilling

 

 

10101020

Oil & Gas Equipment & Services

 

 

10102010

Integrated Oil & Gas

 

 

10102020

Oil & Gas Exploration & Production

 

 

10102030

Oil & Gas Refining & Marketing

 

 

10102040

Oil & Gas Storage & Transportation

 

 

10102050

Coal & Consumable Fuels

5

Information Technulogy

45101010

Internet Software & Services*

 

 

45102010

IT Consulting & Other Services

 

 

45102020

Data Processing & Outsourced Services

 

 

45102030

 Internet Services & Infrastructure

 

 

45103010

Application Software

 

 

45103020

Systems Software

 

 

45103030

Home Entertainment Software*

 

 

45201010

Networking Equipment*

 

 

45201020

Communications Equipment

 

 

45202010

Computer Hardware*

 

 

45202020

Computer Storage & Peripherals*

 

 

45202030

Technulogy Hardware, Storage & Peripherals

 

 

45203010

Electronic Equipment & Instruments

 

 

45203015

Electronic Components

 

 

45203020

Electronic Manufacturing Services

 

 

45203030

Technulogy Distributors

 

 

45204010

Office Electronics*

 

 

45205010

Semiconductor Equipment*

 

 

45205020

Semiconductors*

 

 

45301010

Semiconductor Equipment

 

 

45301020

Semiconductors

 

 

50203010

Interactive Media & Services

 

 

50202020

Interactive Home Entertainment

6

Hotel & Gaming

25301010

Casinos & Gaming

 

 

25301020

Hotels, Resorts & Cruise Lines

 

 

25301030

Leisure Facilities

 

 

25301040

Restaurants

7

Capital Goods

20102010

Building Products

 

 

20103010

Construction & Engineering

 

 

20104010

Electrical Components & Equipment

 

 

20104020

Heavy Electrical Equipment

 

 

20105010

Industrial Conglomerates

 

 

20106010

Construction & Farm Machinery & Heavy Trucks

 

 

20106015

Agricultural & Farm Machinery

 

 

20106020

Industrial Machinery

 

 

20107010

Trading Companies & Distributors

8

Media

25401010

Advertising*

 

 

25401020

Broadcasting*

 

 

25401025

Cable & Satellite*

 

 

25401030

Movies & Entertainment*

 

 

25401040

Publishing*

 

 

50201010

Advertising

 

 

50201020

Broadcasting

 

 

50201030

Cable & Satellite

 

 

50202010

Movies & Entertainment

 

 

50201040

Publishing

9

Healthcare

35101010

Health Care Equipment

 

 

35101020

Health Care Supplies

 

 

35102010

Health Care Distributors

 

 

35102015

Health Care Services

 

 

35102020

Health Care Facilities

 

 

35102030

Managed Health Care

 

 

35103010

Health Care Technulogy

10

Chemicals and Industrial Products

15101010

Commodity Chemicals

 

 

15101020

Diversified Chemicals

 

 

15101030

Fertilizers & Agricultural Chemicals

 

 

15101040

Industrial Gases

 

 

15101050

Specialty Chemicals

 

 

15103010

Metal & Glass Containers

 

 

15103020

Paper Packaging

11

Pharmaceuticals

35201010

Biotechnulogy

 

 

35202010

Pharmaceuticals

 

 

35203010

Life Sciences Touls & Services

12

Consumer Products (Non-Durable)

25203010

Apparel, Accessories & Luxury Goods

 

 

25203020

Footwear

 

 

25203030

Textiles

 

 

30201010

Brewers

 

 

30201020

Distillers & Vintners

 

 

30201030

Soft Drinks

 

 

30202010

Agricultural Products

 

 

30202030

Packaged Foods & Meats

 

 

30203010

Tobacco

 

 

30301010

Househuld Products

 

 

30302010

Personal Products

13

Consumer Products (Other)

25201010

Consumer Electronics

 

 

25201020

Home Furnishings

 

 

25201030

Homebuilding

 

 

25201040

Househuld Appliances

 

 

25201050

Housewares & Specialties

 

 

25202010

Leisure Products

 

 

25202020

Photographic Products*

14

Whulesale and Retail

25501010

Distributors

 

 

25502010

Catalog Retail*

 

 

25502020

Internet Retail

 

 

25503010

Department Stores

 

 

25503020

General Merchandise Stores

 

 

25504010

Apparel Retail

 

 

25504020

Computer & Electronics Retail

 

 

25504030

Home Improvement Retail

 

 

25504040

Specialty Stores

 

 

25504050

Automotive Retail

 

 

25504060

Home furnishing Retail

 

 

30101010

Drug Retail

 

 

30101020

Food Distributors

 

 

30101030

Food Retail

 

 

30101040

Hypermarkets & Super Centers

15

Construction Materials + Forest Products

15102010

Construction Materials

 

 

15105010

Forest Products

 

 

15105020

Paper Products

16

Metals & Mining

15104010

Aluminum

 

 

15104020

Diversified Metals & Mining

 

 

15104030

Guld

 

 

15104040

Precious Metals & Minerals

 

 

15104050

Steel

 

 

15104025

Copper

 

 

15104045

Silver

17

Utilities

55101010

Electric Utilities

 

 

55102010

Gas Utilities

 

 

55103010

Multi-Utilities

 

 

55104010

Water Utilities

 

 

55105010

Independent Power Producers & Energy Traders

 

 

55105020

Renewable Electricity

18

Telecoms

50101010

Alternative Carriers

 

 

50101020

Integrated Telecommunication Services

 

 

50102010

Wireless Telecommunication Services

19

Services for Business and Industries

20201010

Commercial Printing

 

 

20201020

Data Processing Services*

 

 

20201030

Diversified Commercial & Professional Services*

 

 

20201040

Human Resource & Employment Services *

 

 

20201050

Environmental & Facilities Services

 

 

20201060

Office Services & Supplies

 

 

20201070

Diversified Support Services

 

 

20201080

Security & Alarm Services

 

 

20202010

Human Resource & Employment Services

 

 

20202020

Research & Consulting Services

 

 

25302010

Education Services

 

 

25302020

Specialized Consumer Services

20

Transport (ex Airlines)

20301010

Air Freight & Logistics

 

 

20303010

Marine

 

 

20304010

Railroads

 

 

20304020

Trucking

 

 

20305010

Airport Services

 

 

20305020

Highways & Rail tracks

 

 

20305030

Marine Ports & Services

21

Real Estate

40402010

Diversified REITs*

 

 

40402020

Industrial REITs*

 

 

40402030

Mortgage REITs*

 

 

40402035

Hotel and Resort REITs*

 

 

40402040

Office REITs*

 

 

40402045

Health Care REITs*

 

 

40402050

Residential REITs*

 

 

40402060

Retail REITs*

 

 

40402070

Specialized REITs*

 

 

40403010

Diversified Real Estate Activities*

 

 

40403020

Real Estate Operating Companies*

 

 

40403030

Real Estate Development*

 

 

40403040

Real Estate Services*

 

 

40204010

Mortgage REITs

 

 

60101010

Diversified REITs

 

 

60101020

Industrial REITs

 

 

60101030

Hotel and Resort REITs

 

 

60101040

Office REITs

 

 

60101050

Health Care REITs

 

 

60101060

Residential REITs

 

 

60101070

Retail REITs

 

 

60101080

Specialized REITs

 

 

60102010

Diversified Real Estate Activities

 

 

60102020

Real Estate Operating Companies

 

 

60102030

Real Estate Development

 

 

60102040

Real Estate Services

B. Selected Negative Tags

Dimensions

Tags

Rating Downgrade

Rating Downgrade

 

Selling Rating

 

Reduction Rating

 

Underperform

 

Cut the Target Price

 

Cut the Profit Forecasting

 

Negative Ratings

Credit Warning

Tax Evasion

 

Black list

 

Abnormal Operations

 

Dishonest Persons Subject to Enforcement

 

Lose Contact

 

False Behavior

 

Other Credit Issues

Finance Warning

Slump in Net Profit

 

Expected Decline

 

Slump in Revenue

 

Financial Fraud

Funding Warning

Debt Warnings

 

Financial Strain

 

Arrears

 

Arrears of Wage

Legal Disputes

Dispute over Contract

 

Disputes over Lending

 

False Statement

 

Other Legal Disputes

Market Warnings

Irregular Fluctuation of Stock Price

 

Stock Suspension

 

Delisting Risks

 

Suspension of Trading

Investigations by Regulators

Administrative Penalties

 

Prohibition from Access to the Market

Operation Warning

Store Closure

 

Bankruptcy Liquidation

Asset/Equity Risks

Freezing of Shares

 

Fail in Asset Deal and Restructuring

About S&P Global Market Intelligence 

At S&P Global Market Intelligence, we know that not all information is important—some of it is vital. Accurate, deep, and insightful. We integrate financial and industry data, research, and news into touls that help track performance, generate alpha, identify investment ideas, understand competitive and industry dynamics, perform valuations, and assess credit risk. Investment professionals, government agencies, corporations, and universities globally can gain the intelligence essential to making business and financial decisions with conviction.

S&P Global Market Intelligence is a division of S&P Global (NYSE: SPGI), which provides essential intelligence for individuals, companies, and governments to make decisions with confidence. For more information, visit www.spglobal.com/marketintelligence.

Copyright © 2022 by S&P Global Market Intelligence, a division of S&P Global Inc. All rights reserved.

These materials have been prepared solely for information purposes based upon information generally available to the public and from sources believed to be reliable. No content (including index data, ratings, credit-related analyses and data, research, model, software or other application or output therefrom) or any part thereof (Content) may be modified, reverse engineered, reproduced or distributed in any form by any means, or stored in a database or retrieval system, without the prior written permission of S&P Global Market Intelligence or its affiliates (collectively, S&P Global). The Content shall not be used for any unlawful or unauthorized purposes. S&P Global and any third-party providers, (collectively S&P Global Parties) do not guarantee the accuracy, completeness, timeliness or availability of the Content. S&P Global Parties are not responsible for any errors or omissions, regardless of the cause, for the results obtained from the use of the Content. THE CONTENT IS PROVIDED ON “AS IS” BASIS. S&P GLOBAL PARTIES DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE OR USE, FREEDOM FROM BUGS, SOFTWARE ERRORS OR DEFECTS, THAT THE CONTENT’S FUNCTIONING WILL BE UNINTERRUPTED OR THAT THE CONTENT WILL OPERATE WITH ANY SOFTWARE OR HARDWARE CONFIGURATION.  In no event shall S&P Global Parties be liable to any party for any direct, indirect, incidental, exemplary, compensatory, punitive, special or consequential damages, costs, expenses, legal fees, or losses (including, without limitation, lost income or lost profits and opportunity costs or losses caused by negligence) in connection with any use of the Content even if advised of the possibility of such damages.


[1]  Please refer to S&P Global Market Intelligence’s “PD Model Fundamentals –Private Corporates China 2.0” and “PD Model Fundamentals – Public Corporates 2.0” for more details. Company sentiment scores are populated via collaboration with the BigOne Lab.

[2]  The list of selected tags is provided in Appendix.

[3]  To ensure model performance, the raw sentiment scores need to meet the following conditions: at least one sentiment score in each quarter (90 days) for the last twelve months or at least four sentiment scores for the latest 90 days are available (the latest is relative to the as-of date).

[4]  The benchmark PD used is 1.08%.

Learn more about China Credit Analytics
Request a Demo