Guide to AVMs

AVMs – Increasing levels of confidence

Guide to AVMs

Automated valuation models have now been around in the UK for nearly 10 years, and with more than 90% of lenders by market share in this country using some form of AVM within their organisation, it is reasonable to assume that the reader does not wish to digest yet another article on the benefits of the AVM. In this briefing, I therefore wanted to begin the debate about the future developments of AVMs where it is envisaged they will not just provide the price of a property at a given point in time, but that they will play a significant role in risk mitigation, fraud prevention, identification of marketing opportunities and not just on a real time basis but with a sound predictive methodology.

AVMs were greeted with initial scepticism and only moved up the lender list of priorities in the days when HIPs were originally envisaged to include a Home Condition Report – with the most cost effective and speedy solution to provide the valuation of the property in question, being an AVM. However, even without mandatory HCRs, lenders are now taking advantage of this technological advancement and have adapted it for other uses such as point of sale decisions, and regular assessment of the value of a loan portfolio at any time for a variety of purposes.

Most lenders are now very comfortable with the concept of confidence levels and appreciate that they genuinely reflect the statistical accuracy of the valuation. They are therefore used as the primary measure by which the suitability of the AVM output is gauged. Setting robust business rules within the AVM is a continuation of this adoption process, and each lender’s approach is unique as they tailor the AVM to reflect their lending decisioning. This is even more important since the arrival of the POSO. Setting automated business rules continues to speed up the process, and again removes what can otherwise be cumbersome and highly manual tasks. It can be as simple as the lender wishes such as ‘only use for remortgages not purchases’ or ‘only accept the AVM if the confidence level is high’, or even more complex by including a combination of rules. Lenders may accept even lower confidence levels with low LTVs or they may simply require a valuation returned regardless of the confidence level when lending to their ‘platinum customers’ after pre credit scoring.

There is also another level to setting business rules and this entails linking into cascading product systems so that the lender can access multiple AVMs, possibly cascading them based on accuracy, price or any given rule set. Several lenders already choose to have more than one AVM provider but currently the decision around which AVM is used, does not make use of advanced intelligent decisioning rules, however this should change in the near future.

In addition to the above, it is understandable that lenders would like to be clear about the risks associated with valuations whether from a physical inspection or automated, because traditionally the output is in the form of ‘spot valuations’, i.e. the valuation process will return a figure as at a given point in time. It would be more valuable for the lender if the valuation process could plot historic and predict future price trends for specific properties thereby indicating whether the subject property is on the increase, decrease or is likely to remain stable.

Another stage in the AVM evolution has been to give lenders the choice of taking associated products and/or given control over the valuation and associated output i.e. users will be able to choose the type of AVM they need, the level of information returned, the level of accuracy they require, the level of automation and/or integration with the supplier and the price will vary accordingly with these choices made. Imagine therefore the ability to move beyond a ‘simple’ AVM and go for a ‘shopping basket’ approach to add products such as:

• reinstatement values, which can enable point of sale buildings insurance,

• rental information can bring the benefits of the AVM to the ever expanding Buy to Let market

• an insurance policy which can mitigate risk in the event of an inaccurate valuation at the time of the application being identified when a lender is forced to exercise their power of sale.

Mortgage fraud, in its many forms, is ever present in our industry and some reports suggest that it is on the increase. Many lenders have already turned to technology which is often embedded into the loan approval process to verify the borrower’s identity, pick up multiple applications etc, but what they may not be aware of are the various fraud prevention tools that are currently available to them as AVM users:

• The ability to alert the user when the subject property attributes they have been provided with, do not match the information stored on record, e.g. the property held on file has 2 bedrooms and not 4 as the user is suggesting. This is not to say that just because the information conflicts, the applicant data is incorrect, but a lender may in these circumstances wish to have a policy rule that seeks a physical inspection if there is ambiguity.

• The address matching process is also able to validate that the property is residential as oppose to commercial.

• Lenders often ask the question, ‘how do we know the property actually exists?’ The response to this is that the AVM can be set to not return a value unless the property can be identified. The address data is derived from PAF (Postal Address File), and is updated monthly providing good validation that the property exists, aerial photography is beginning to help here too. What is fundamental however is that as AVMs evolve further, solutions to overcome these issues are being developed and enhanced.

When considering AVMs as an additional tool to assist lenders with fraud prevention in the future, it is worth taking a look at how it is already being managed elsewhere in a global market, and UKValuation is well placed to introduce fraud preventative technology to the UK as a result of the vast experience of its American parentage. First American CoreLogic, a member of The First American Corporation family of companies, is the largest provider of AVMs in the US and has been instrumental in helping lenders in the US market successfully tackle fraud issues. Risk associated with mortgage lending has traditionally been managed through labour intensive quality control and due diligence reviews, however, the combined resources now available to the US market makes this process more efficient and effective, with the business being able to apply advanced, unique data and predictive analysis resources throughout the mortgage lending process. There are 3 broad areas of fraud detection technologies:

i) Data validation compares key elements from the application with public records and credit data information to verify accuracy.

ii) A property address can be checked against a detailed history of the neighbourhood and prior sales to locate instances of property overvaluation and flipping (buying and quickly reselling properties for large profits, usually after speedy renovations). Once the tool has identified any issues with the neighbourhood, it can then identify the level of valuation risk exposure and the appropriate method to employ from a physical inspection, through to drive-by, and AVM.

iii) Fraud pattern recognition is where a statistical model will find hidden patterns of fraud in the data and will return a ’fraud score’, as the data provided will be checked against known fraud patterns.

In summary, the future of the automated valuation model is clearly not just about speeding up the lending process, reducing costs and improving accuracy, these are expected by lenders as a matter of course and over time, these three variables will reach their optimum level. What will be of interest, particularly as new AVM providers come to the market is what they can ‘bring to the party’ because in order to satisfy the challenges that their potential clients will require to be overcome, it will need to be more than just an AVM.