The global market for telecommunications is expanding rapidly. Increasing usage patterns combined with declining revenue underline the ongoing price erosion caused by a highly competitive market. Thus, efficient revenue streams are critical to the financial wellbeing of a telecommunications operator.

As with any organisation offering ‘post-pay’ services, telecommunications carries certain risks. The customer’s use of the services is instantaneous; however, the payment for those services lags behind the usage by some time. How to estimate future non-performing assets and having adequate levels of provisions against these has an impact on the volatility and cyclicality of telecom companies’ profits and their cost of capital.

Organisations are under pressure to identify accurate provisions for anticipated bad debt in order to effectively manage risks. The majority of the large organisations either base their provisions on established loss forecasting techniques or they use loss forecasting techniques to extend their default probability calculations from a one year window into a 3-5 year period. Changes in acquisition policies and procedures, and national and local economic and business conditions should be factored in during loss forecasting.

Understanding the different methods with which to forecast losses, and knowing when to use each method, allows for greater control of portfolio management and more profitable account management strategies. This article discusses the following loss forecasting methodologies that could be used in estimating future non-performing assets:

• Roll rate-based forecasting
• Vintage-based forecasting
• Score-based forecasting
• Markov Chains

Roll rate-based forecasting

Using historical and actual data, outstandings are calculated on a monthly basis by delinquency level. Using this data, monthly roll rates are calculated for each delinquency level. Forecasted roll rates are calculated as a weighted average of last six months’ average roll rate and seasonality (Sep/Oct in the table below). Calculated roll rates are multiplied with the current month’s outstandings to obtain forecasted values.

The strengths of this approach are that the forecasts are accurate for 9-12 months, it is flexible (bankruptcies or special arrangement accounts could be separately forecasted), plus it can offer insight into where potential problems might be. However, considerations that must be taken into account are that this methodology requires an independent outstandings forecast (generally from the finance department) and that it ignores local and global economic factors, recent policy and procedure changes, and the quality and characteristics of recent loans.

Vintage-based forecasting

‘Vintage’ refers to opening date; vintage-based loss forecasts work on the assumption that recent accounts will mature in a similar way to the older accounts. Losses by month over the life of older accounts are calculated and loss curves are developed showing the percentage of losses occurring monthly. Newer accounts are presumed to follow similar loss patterns, thus completing the forecast.

Strengths of this approach are that, again, forecasts are accurate for 9-12 months and that it is a flexible methodology. Vintage-based forecasting also takes into account recent policy changes and economic factors. Considerations here include the requirement of an independent outstandings forecast, and that this approach is also susceptible to influences that could cause a timing shift such as ‘no payment’ promotions.

Score-based forecasting

Score-based forecasts use a percentage of actual losses by score interval for a specific time period, such as 12 or 24 months. Acquisition score is used for new accounts, behavioural score is used for mature accounts. Expected loss rates for 12 months and 24 months respectively by specific score ranges are calculated. Then, adjustments are made to the 24 month loss figure, where it reflects losses that occur between months 13 and 24. After determining the current outstandings distribution by the same score ranges, the expected losses for year 1 and year 2 are calculated by multiplying the expected loss rates in each score range by the outstandings that sit in that score range.

Score-based forecasting has the strengths of being accurate for the specified time period, it is helpful for brand new segments, and above all, it is an easy methodology to use. Considerations that must be made include the fact that an effective time period depends on the observation period used during the score development, a population shift and score deterioration could impact accuracy, and again, this approach ignores local and global economic factors.

Markov Chains

Markov Chains based loss forecasting methodology requires the creation of a transition matrix, which indicates the probability of moving from one state to another during a specific timeframe.

To calculate losses, various delinquency states (including charge-offs) of the portfolio are determined and the transition matrix is created. Then, the transition matrix is multiplied by itself to provide probabilities of movement between the states for specific periods into the future. As the final step, the final matrix is multiplied by the portfolio outstandings distribution to give forecasted losses.

Finally, the strengths of the Markov Chain methodology are that it provides accurate and easy forecasts for 12+ months and it takes into account both delinquency improvements and deteriorations within the portfolio. The Consideration here is that time periods to develop the transition matrix need to be selected very carefully. The use of long-term averages and stress test would address this concern. This approach also ignores recent policy and procedure changes and the quality and characteristics of recent loans. However, calibration of the transition matrix would alleviate this issue.

Experian Decision Analytics is an expert in providing data, analytics, model development, reporting and implementation software. With its time-tested and proven analytical capabilities, it can help organisations develop and seamlessly integrate adequate and effective loss forecasting and provisioning methodologies into their credit risk management environment.

Burak Kilicoglu – Business Consultant, Experian Decision Analytics – From his presentation given at the Experian Decision Analytics Telco Forum 2007

Contact us for further discussions about this article

View this article in German

Top