UK: IFRS 9 And The "Is It A Bird, Or Is It A Bomber?" Problem

Last Updated: 23 February 2017
Article by Alexander Marianski and Ian Wilson

Most Read Contributor in UK, August 2017

With the adoption of the IFRS 9 accounting standard into EU law, it is full steam ahead for banks to deploy credit models that estimate Expected Credit Loss (ECL) accounting values. The standard requires firms to account for lifetime ECL on loans that have experienced a "significant increase in credit risk" (SICR), but allows firms to reach their own conclusions as to just how much credit risk ought to be viewed as "significant".

Most firms have developed their first suite of ECL models, and are able to estimate 12-month and lifetime ECL under various input assumptions such as forward economic scenarios. However, many firms have held off finalising their definition of SICR until additional forward economic scenarios have been developed. The wait-and-see approach also allows some degree of consensus in interpretation to emerge.

Credit risk, and movements in credit risk, are of course quantifiable using well-established techniques. But what should count as a "significant" increase? The IFRS 9 standard strongly suggests in paragraph 5.5.9 that SICR should be a function of movements in remaining lifetime PD as well as other qualitative factors, but gives little detail on how or where to calibrate the PD threshold.

In this article, we explore binary classifier techniques borrowed from machine learning applications, and explain how they can be used to select the optimal definition of SICR.

Receiver Operating Characteristic

Pioneers of Radar faced a similar problem in choosing the threshold at which to call the blip on the screen a big flock of birds, or a bomber. The Receiver Operating Characteristic (ROC) curve, which measures the discriminatory power of a binary classifier and is familiar to scorecard modellers, has its roots in radar engineering.

With the application of machine learning techniques for automated decision-making, it became necessary to set a decision rule which optimises the trade-off between identifying all bombers (the true positive result), all birds (the true negative result) and falsely calling bombers birds (false negative) or birds bombers (false positive).

Other examples of automatic classifiers include sorting of post, recognising hand-written numbers on cheques, and medical imaging. Many applications introduce the concept of a "cost function" – in the postal example the cost of misclassification is empirically measurable in terms of additional time spent resolving errors as well as compensation to customers.

IFRS 9 features and cost

The science of machine learning places considerable emphasis on first identifying which characteristics are of interest – termed the "feature vector" – as well as the relative cost of correct and incorrect classification. The IFRS 9 standard allows us to circumvent this first step by providing a handy (and concise) non-exhaustive list of factors which may be included in the feature vector (paragraph B5.5.17) including changes in credit spread, credit rating, economic conditions, borrower's financial results, and pricing.

In addition, the IFRS 9 standard disallows losses (i.e. the inclusion of collateral cash flows) as an input into stage allocation (paragraph B5.5.9), leading us to conclude that expected credit losses (i.e. the obvious choice of cost function) should not be introduced to weight the classification to stage 1 or 2. Based on this information, standard machine learning techniques can be used to set the optimal boundary for classifying accounts into those likely to default (which should be classed as stage 2) and those not (which belong in stage 1).

Linear Discriminant Functions

In the case where the classes are linearly separable and assumed multivariate Normal, then the straight line through the origin which gives maximum class separation can be found with relative ease to be a linear combination of feature vectors on the margin or, equivalently, a function of the class means and standard deviations. The use of a discriminant function handily avoids the problem of having to estimate probability densities, and the approach can be extended to:

  • Situations where the classes are not separable; and
  • Non-linear boundaries between classes, as alluded to by the IFRS 9 standard (paragraph B5.5.9) permitting a larger absolute movement for the best-rated customers at initial recognition.

The four possible results of a binary classifier are:

  • True positive;
  • True negative;
  • False positive; and
  • False negative.

These can be described using a 2x2 matrix, often referred to as a confusion matrix. Note that when weighted by a cost function, the term "cost matrix" is also used. Many market participants have adopted approaches which evaluate variations in true positive rate (TPR), true negative rate (TNR), false positive rate (FPR) and false negative rate (FNR). Definitions of "positive" and "negative" are typically articulated using business rules which often include progression to arrears and default from stages 1 and 2, as well as rules around accounts' propensity to flip-flop between stages.

Unsurprisingly, stages 1 and 2 are not separable (an expected result – with perfect separation, banks could choose to lend money only to borrowers who are known never to default) and some trade-off between the four metrics described above is therefore required. In the sciences, Matthews Correlation Coefficient and Mean Square Error are often used to combine the metrics into a single criterion to be minimised. However, few banks have combined the four metrics, and have instead relied on judgement to select their preferred trade-off between business rules, i.e. a semi-informed guess at the optimal boundary between stage 1 and stage 2.

Perceptron and Multi-Layer Networks

With a linear decision boundary, this approach is consistent with analysis of linear discriminants. However, the decision boundary is often non-linear and the Normality assumption mentioned is usually violated. Linear discriminant analysis can be extended to cope with non-linear boundaries by transforming explanatory variables before taking the weighted combination. This is the "perceptron". The perceptron permits a much wider range of decision boundaries, and is trained using gradient descent. Since the confusion matrix may have regions of constant performance, perceptron training instead introduces an error function which always has non-zero gradient.

The perceptron is, however, not able to deal with any general separation problem such as SICR. A perhaps extreme example of a problem which cannot be solved using a single perceptron is the "exclusive or" problem. A generalised decision boundary can, however, be achieved by introducing successive transformations or "layers". This is the "multi-layer perceptron", and is the cornerstone of modern artificial intelligence. A two-layer network is sufficient to approximate any continuous decision boundary.


The general problem of how to find the boundary between stages 1 and 2 can be addressed by introducing "features" described in the IFRS 9 standard as a vector of information, and training a multi-layer perceptron to discriminate between accounts which subsequently default, and accounts which do not.

It is, however, important to recognise that:

  • When a machine "learns" to mimic observed patterns of behaviour, it has not learned to articulate the rationale for its results. Therefore, the outputs of a multi-layer perceptron (and non-linear decision boundaries) can be difficult or impossible for management to explain.
  • Monotonicity seems likely to be a desirable feature of decision boundaries in credit risk applications. Therefore, it is likely that sufficient SICR classification performance can be achieved using a single layer perceptron.
  • It may be desirable to enforce an intuitive functional form on the decision boundary, such a "PD multiple" (straight line through the origin) or "rating notch difference" (probit or logit). This can in fact be achieved without the perceptron: Linear discriminant techniques also work if features are pre-mapped into a higher-dimensional space using the desired functional form.

A final word of caution: This kind of analysis is data-intensive and should be performed across the economic cycle. After all, the radar operator who calibrates the receiver in December may be in for a shock when birds return from their winter migration.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.

To print this article, all you need is to be registered on

Click to Login as an existing user or Register so you can print this article.

In association with
Related Video
Up-coming Events Search
Font Size:
Mondaq on Twitter
Register for Access and our Free Biweekly Alert for
This service is completely free. Access 250,000 archived articles from 100+ countries and get a personalised email twice a week covering developments (and yes, our lawyers like to think you’ve read our Disclaimer).
Email Address
Company Name
Confirm Password
Mondaq Topics -- Select your Interests
 Law Performance
 Law Practice
 Media & IT
 Real Estate
 Wealth Mgt
Asia Pacific
European Union
Latin America
Middle East
United States
Worldwide Updates
Check to state you have read and
agree to our Terms and Conditions

Terms & Conditions and Privacy Statement (the Website) is owned and managed by Mondaq Ltd and as a user you are granted a non-exclusive, revocable license to access the Website under its terms and conditions of use. Your use of the Website constitutes your agreement to the following terms and conditions of use. Mondaq Ltd may terminate your use of the Website if you are in breach of these terms and conditions or if Mondaq Ltd decides to terminate your license of use for whatever reason.

Use of

You may use the Website but are required to register as a user if you wish to read the full text of the content and articles available (the Content). You may not modify, publish, transmit, transfer or sell, reproduce, create derivative works from, distribute, perform, link, display, or in any way exploit any of the Content, in whole or in part, except as expressly permitted in these terms & conditions or with the prior written consent of Mondaq Ltd. You may not use electronic or other means to extract details or information about’s content, users or contributors in order to offer them any services or products which compete directly or indirectly with Mondaq Ltd’s services and products.


Mondaq Ltd and/or its respective suppliers make no representations about the suitability of the information contained in the documents and related graphics published on this server for any purpose. All such documents and related graphics are provided "as is" without warranty of any kind. Mondaq Ltd and/or its respective suppliers hereby disclaim all warranties and conditions with regard to this information, including all implied warranties and conditions of merchantability, fitness for a particular purpose, title and non-infringement. In no event shall Mondaq Ltd and/or its respective suppliers be liable for any special, indirect or consequential damages or any damages whatsoever resulting from loss of use, data or profits, whether in an action of contract, negligence or other tortious action, arising out of or in connection with the use or performance of information available from this server.

The documents and related graphics published on this server could include technical inaccuracies or typographical errors. Changes are periodically added to the information herein. Mondaq Ltd and/or its respective suppliers may make improvements and/or changes in the product(s) and/or the program(s) described herein at any time.


Mondaq Ltd requires you to register and provide information that personally identifies you, including what sort of information you are interested in, for three primary purposes:

  • To allow you to personalize the Mondaq websites you are visiting.
  • To enable features such as password reminder, newsletter alerts, email a colleague, and linking from Mondaq (and its affiliate sites) to your website.
  • To produce demographic feedback for our information providers who provide information free for your use.

Mondaq (and its affiliate sites) do not sell or provide your details to third parties other than information providers. The reason we provide our information providers with this information is so that they can measure the response their articles are receiving and provide you with information about their products and services.

If you do not want us to provide your name and email address you may opt out by clicking here .

If you do not wish to receive any future announcements of products and services offered by Mondaq by clicking here .

Information Collection and Use

We require site users to register with Mondaq (and its affiliate sites) to view the free information on the site. We also collect information from our users at several different points on the websites: this is so that we can customise the sites according to individual usage, provide 'session-aware' functionality, and ensure that content is acquired and developed appropriately. This gives us an overall picture of our user profiles, which in turn shows to our Editorial Contributors the type of person they are reaching by posting articles on Mondaq (and its affiliate sites) – meaning more free content for registered users.

We are only able to provide the material on the Mondaq (and its affiliate sites) site free to site visitors because we can pass on information about the pages that users are viewing and the personal information users provide to us (e.g. email addresses) to reputable contributing firms such as law firms who author those pages. We do not sell or rent information to anyone else other than the authors of those pages, who may change from time to time. Should you wish us not to disclose your details to any of these parties, please tick the box above or tick the box marked "Opt out of Registration Information Disclosure" on the Your Profile page. We and our author organisations may only contact you via email or other means if you allow us to do so. Users can opt out of contact when they register on the site, or send an email to with “no disclosure” in the subject heading

Mondaq News Alerts

In order to receive Mondaq News Alerts, users have to complete a separate registration form. This is a personalised service where users choose regions and topics of interest and we send it only to those users who have requested it. Users can stop receiving these Alerts by going to the Mondaq News Alerts page and deselecting all interest areas. In the same way users can amend their personal preferences to add or remove subject areas.


A cookie is a small text file written to a user’s hard drive that contains an identifying user number. The cookies do not contain any personal information about users. We use the cookie so users do not have to log in every time they use the service and the cookie will automatically expire if you do not visit the Mondaq website (or its affiliate sites) for 12 months. We also use the cookie to personalise a user's experience of the site (for example to show information specific to a user's region). As the Mondaq sites are fully personalised and cookies are essential to its core technology the site will function unpredictably with browsers that do not support cookies - or where cookies are disabled (in these circumstances we advise you to attempt to locate the information you require elsewhere on the web). However if you are concerned about the presence of a Mondaq cookie on your machine you can also choose to expire the cookie immediately (remove it) by selecting the 'Log Off' menu option as the last thing you do when you use the site.

Some of our business partners may use cookies on our site (for example, advertisers). However, we have no access to or control over these cookies and we are not aware of any at present that do so.

Log Files

We use IP addresses to analyse trends, administer the site, track movement, and gather broad demographic information for aggregate use. IP addresses are not linked to personally identifiable information.


This web site contains links to other sites. Please be aware that Mondaq (or its affiliate sites) are not responsible for the privacy practices of such other sites. We encourage our users to be aware when they leave our site and to read the privacy statements of these third party sites. This privacy statement applies solely to information collected by this Web site.

Surveys & Contests

From time-to-time our site requests information from users via surveys or contests. Participation in these surveys or contests is completely voluntary and the user therefore has a choice whether or not to disclose any information requested. Information requested may include contact information (such as name and delivery address), and demographic information (such as postcode, age level). Contact information will be used to notify the winners and award prizes. Survey information will be used for purposes of monitoring or improving the functionality of the site.


If a user elects to use our referral service for informing a friend about our site, we ask them for the friend’s name and email address. Mondaq stores this information and may contact the friend to invite them to register with Mondaq, but they will not be contacted more than once. The friend may contact Mondaq to request the removal of this information from our database.


This website takes every reasonable precaution to protect our users’ information. When users submit sensitive information via the website, your information is protected using firewalls and other security technology. If you have any questions about the security at our website, you can send an email to

Correcting/Updating Personal Information

If a user’s personally identifiable information changes (such as postcode), or if a user no longer desires our service, we will endeavour to provide a way to correct, update or remove that user’s personal data provided to us. This can usually be done at the “Your Profile” page or by sending an email to

Notification of Changes

If we decide to change our Terms & Conditions or Privacy Policy, we will post those changes on our site so our users are always aware of what information we collect, how we use it, and under what circumstances, if any, we disclose it. If at any point we decide to use personally identifiable information in a manner different from that stated at the time it was collected, we will notify users by way of an email. Users will have a choice as to whether or not we use their information in this different manner. We will use information in accordance with the privacy policy under which the information was collected.

How to contact Mondaq

You can contact us with comments or queries at

If for some reason you believe Mondaq Ltd. has not adhered to these principles, please notify us by e-mail at and we will use commercially reasonable efforts to determine and correct the problem promptly.