Pricing of Non-Performing Assets

Document Sample
Pricing of Non-Performing Assets Powered By Docstoc
                                    Contents                                                            1 of 6


                                            Pricing of Non-Performing Assets
               Junni Zhang
               Peking University, Guanghua School of Management, Department of Business Statistics
                and Econo-metrics, 5 Yiheyuan Road, Beijing 100871, P. R. China.

                      In year 1999, four asset management corporations were established in China. Their original
               mission was to acquire non-performing assets from China’s major state-owned commercial banks and
               China Development Bank, and to manage and dispose of these assets, so as to push forward the reform

               and development of state-owned commercial banks and state-owned enterprises. With rich experience
               gained through managing and disposing of policy-related non-performing assets, and accompanying
               the approval and implementation of a series of reform scheme, the asset management corporations
               have been gradually commercialized.
                      Pricing of non-performing assets has always been a core issue that asset management corpora-
               tions face. In disposal of policy-related non-performing assets, in order to reduce the loss of state-owned
               assets and maximize the recovery amount, the asset management corporations need to rationally price
               the non-performing assets. In the process of commercialization, pricing of non-performing assets
               directly determines the competitiveness of asset management corporations.
                      Asset management corporations and the asset evaluation industry have been continuously im-
               proving the process and methods for pricing non-performing assets, but pricing remains a difficult
               issue. Several reasons for this are listed below.
                                   Contents                                                              2 of 6

                  • The original information about non-performing assets comes from the banks. The asset man-
                    agement corporations cannot know how some of the indicators were formed and whether they
                    are accurate.

                  • Because the debtors often take uncooperative stance in due diligence investigation, it is difficult
                    for asset management corporations to gain accurate information.

                  • Due to various reasons such as debtors’ uncooperative stance and missing information, it is very
                    difficult to carry out normal asset evaluation procedures.

                  • Many non-financial factors have a large impact on the disposal price of non-performing assets,

                    such as debtors’ willingness to pay, the negotiation ability of front line disposal personnel, etc.

                     Therefore, it is very valuable to fully utilize the rich experience gained through historical disposal
               of non-performing assets. In a consulting project for one major asset management corporation, we
               designed and implemented a well-rounded online data collection system, and used it to organize the
               historical experience into historical data, and fit a system of statistical models. These models can be
               used to give a reference price for future non-performing assets.

               The Data Mining Methodology

                     We adopt the CRoss Industry Standard Process for Data Mining (CRISP-DM, please refer to
      for more details) which partitions a data mining project into six phases —
               business understanding, data understanding, data preparation, modeling, evaluation and deployment.
               Contents                                          3 of 6


                Figure 1: Phases of the CRISP-DM Process Model
                                   Contents                                                          4 of 6

                  Specifically for this project, the six phases are as follows:

               1. Business understanding. We communicated with experts on asset evaluation and front-line dis-
                  posal personnel, and collected factors that have large impact on disposal price of non-performing
                  assets and that could be potentially measured quantitatively or qualitatively.

               2. Data understanding. We investigated sources of historical data and their quality, and designed
                  and implemented an online data collection system that were used to collect historical data from
                  various branches into a central database.

               3. Data preparation. We cleaned up errors in the data, and extracted variables that could be used

                  to build models for the price of non-performing assets.

               4. Modeling. We built up a system of statistical models for pricing non-performing assets.

               5. Evaluation. We evaluated the performance of the statistical models in estimating the recovery
                  amount and recovery rate of historical assets, and also tested the performance of the models
                  using some newly collected test data.

               6. Deployment. For future non-performing assets, the relevant information can be input into the
                  data collection system, and the statistical models can then be applied to quickly give reference

                                   Contents                                                            5 of 6

               Data Quality Issues And The New Data Collection System

                     Historically, information about non-performing assets was collected using Excel templates. This
               creates a whole bunch of data quality problems. For example, names and addresses could appear in
               numerical variables, a qualitative variable could have many nonstandard values, there could appear
               various kinds of apparent data errors and data conflicts, the numerical variables could take mutually
               inconsistent values, etc. These quality problems often lie in a big part of the data, and can render any
               serious efforts on data analysis fruitless.
                     We therefore need a new data collection system that can check data quality seriously. Also, this

               new system needs to organize information input by personnel across various branches into a central

               database. We did a detailed study of the paper archives of some historically disposed non-performing
               assets, and collaborated with experts on asset evaluation and front-line disposal personnel in identifying
               variables that should be collected. We then spent a lot of efforts in designing comprehensive rules
               that can ensure data quality by checking individual variables and the relationship between variables in
               the same data table and across data tables. We also provided the system administrator with various
               summary tables so that the data collection progress could be closely monitored.

               The System of Statistical Models

                     Using the new online data collection system, we collected data on historically disposed non-
               performing assets from the paper archives. We then used descriptive statistics to explore the rela-
               tionship between the predictors and recovery amount or recovery rate, and collaborated closely with
           T   experts on asset evaluation and front-line disposal personnel in extracting meaningful variables that
                                  Contents                                                          6 of 6

               could be used to build up prediction models.
                     Considering that the variables are not normally distributed and that many variables have fairly
               large percentages of missing values, we used a combination of statistical techniques such as Box-Cox
               transformation and decision trees. We then selected the best models using a combination of various
               model evaluation statistics.
                     We evaluated extensively the performance of the statistical models in estimating the recovery
               amount and recovery rate, using both historical assets and newly collected test data. Results show
               that the statistical models way outperformed the hypothetical liquidation method, a most-commonly
               adopted accounting-based method for pricing non-performing assets, and could offer reasonable refer-

               ence prices for future non-performing assets.


                     In this project, for the first time in China, we systematically collected large-scale historical
               data on non-performing assets, and established the paradigm of applying data mining to pricing non-
               performing assets. As any data mining project, the resulting data processing procedure and statistical
               models need continuous improvement. New data need to be continuously added to the database, and
               the models need to be updated when the performance deteriorates.


Shared By:
Description: Pricing of Non-Performing Assets