Introducing our Next Generation Scorecard.
After 3 years (and 300,000 applications to date) we continue to demonstrate our commitment to innovation, and creating value for both Lenders and Borrowers, with the launch of our next generation scorecard. Using a combination of data science and machine learning, we have completed a 12 month project that takes big data (credit data, banking and demographic borrower data) to create a new generation credit scoring model, known as scorecard 1.5
The model development involves rigorous testing and analysis, before being vetted and endorsed by Harmoney’s Credit Committee and Board. Then our wholesale funding partners completed their own due diligence.
Some of the analysis completed to test the efficiency of the scoring model is shared below. The results are globally competitive and should see Harmoney leading New Zealand in pricing consumer credit risk.
But we’re not sitting still, as work has already started on scorecard 2.0.
Our first-generation scorecard has performed well.
We launched with New Zealand’s first peer-to-peer lending license in September 2014. Prior to launch, Harmoney had already gained a significant endorsement of our credit model by securing committed investment before facilitating our first loan on the platform – a world first. In fact, Harmoney secured $100m in lending prior to launch from local banks and large investment funds from London and New York to fund loans alongside NZ retail Lenders such as yourself.
We launched to provide a better way to invest and borrow believing in interest rates priced fairly to reflect underlying credit risk. From day one Harmoney has publicly priced 30 risk grades and disclosed a full suite of marketplace statistics, including default history by grade.
It’s time for a change.
From the day a traditional scorecard is launched it will naturally degrade in predictive ability over time. To illustrate this, compare the Auckland housing market today to 3 years ago; people who can afford to buy in the Auckland market today looks different to 3 years ago. So, a scorecard which uses home ownership as a major influencer would be less accurate today.
To address this, our next generation scorecard will utilise machine learning in order to constantly refine and adapt to the changing macro-economic environment in which we operate.
So, what specifically will be changing for Lenders?
- From 3 August 2017, all new loan applications will be assessed with the new generation scorecard
- The scorecard will increase the accuracy of pricing risk significantly from the first generation scorecard
- The probability of a default for each grade will change with the new scorecard.
- Borrower interest rates will reduce across risk grades A1-F5. The minimum rate will be 6.99% p.a. and maximum rate will be 29.99% p.a. This reflects the underlying risk and improved ability to price risk.
- The maximum loan limits are being increased in some grades. However, it is important to note that the affordability testing that is done will not change.
- There will be a 60 day period where the platform will list some loan applications that have been assessed and priced using the original scorecard and pricing model as well as some other loan applications that have been assessed and priced using the new scorecard and pricing model. This is due to approvals provided to borrowers before 3 August 2017 being valid for up to 60 days.
What do you need to do?
- Satisfy yourself that Lending under the new parameters is right for you. Keep in mind that credit scoring uses historical data to produce a credit model, together with information about a specific Borrower that is current when they make their application is used to assess the credit score for that Borrower. Ultimately, you will be lending to a human being and there is always the risk of their circumstances changing.
- Harmoney provides a detailed risk section. Consider the risks associated with Lending prior to committing any funds on the platform.
- Review the changes outlined and consider if Lending via the Harmoney platform continues to meet your needs. Note the changes to pricing and to the estimate of default probability.
- Auto-lend will be deactivated for all lenders when these changes come into effect. If you have Auto-lend activated you will need to re-activate that function in your lender dashboard if you wish to auto-lend funds under the new parameters.
- These changes apply to new lending only. There will be no changes to the existing loans in your portfolio.
The new borrower scorecard in more detail.
We regularly review our policy and process with regards to underwriting risk. Harmoney has a data science team tasked with producing analytical work to improve the quality of the decisions Harmoney makes in the approval and risk grading of potential borrowers who list their loans on the marketplace. You can read more about changes to risk underwriting process or policy here.
The scorecard (short name for ‘credit scoring predictive model’) is an important part of the process of underwriting risk. It takes borrower data and runs it through an algorithm to give each borrower/s a credit grade and associated interest rate and probability of default. The current borrower scorecard (1.0), which has been in place since Harmoney launched in September 2014, is being replaced by a new scorecard (1.5) to provide a greater degree of accuracy in determining the borrowers risk grade. We have taken loan performance data from over 20,000 loans and used machine learning to create this improved scorecard.
The scorecard assigns each borrower a probability of default (pD) and an interest rate which is shown as a risk grade. The new scorecard means that the pD and interest rate for each risk grade has been reset. This means that a borrower that was a C3 on the old scorecard is unlikely to be a C3 on the new scorecard.
A standard test for any credit scoring model is the Gini coefficient score. Once the model is ready for testing actual loan performance data is used to test how well the scorecard can predict credit defaults from borrowers.
The redline on the graph would be as accurate as flipping a coin. The Gini score measures the difference between flipping a coin and 100% accuracy in predicting future credit risk.
Scorecard 1.5 has a Gini of 74.1% whereas Scorecard 1 has a Gini of 53.4% - in other words, scorecard 1.5 improves predictive accuracy by 38%.
Underpinning Harmoney’s ability to make such improvements is the sheer volume of data collected over the last three years (more detail provided below).
A Gini Coefficient is a scale of predictive power from 0 to 1. A higher Gini means more predictive power, a lower Gini means less predictive power. A Gini of 0 is the statistical equivalent of a coin toss – those who score high are no statistically more likely to repay than those who score low. A Gini of 1 is perfectly predictive – the score is able to identify who will repay and who will default with 100% accuracy every time.
Updating borrower interest rates, loan limits, and probability of default.
With the introduction of the the new scorecard borrower interest rates, loan limits and default rates are being updated.
Updating Borrower interest rates.
The new scorecard means that all risk grades repriced with a new interest rate, with a minimum rate of 6.99% and a maximum interest rate of 29.99%, effective 3 August 2017. The new scorecard has recalibrated the risk grades and the type of Borrower in each grade. This combined with the improved accuracy of the new scorecard providing greater certainty that borrowers have been graded correctly (see the section on the Gini) means that interest rates can be sharpened. The new lower interest rates should result in a higher volume of loans available to lenders for investment.
Increasing borrower loan limits in some grades.
The maximum amount that can be borrowed in each risk grade is also changing with the maximum limit for each grades ranging from $10,000 for F-grade borrowers to $70,000 for A-grade borrowers. As these limits are maximum limits per grade- the loan amount borrowers will be approved for is based on their affordability to repay the loan with the maximum amount being the grade limit.
Update to the annual Probability of default (pD) for risk grades.
The probability of default rate (pD) is the forecast probability that a loan will default with no further repayments are expected to be received. A different rate is assigned to each risk grade. The PD rate is shown as an annual rate to enable comparison to the annual interest rate (you can learn more about PD rates here).
The annual pD rates have been recalibrated to the forecast default rates for each grade with the new scorecard.
|Risk grade||Borrower Interest Rate p.a.||Forecast Probability of Default p.a.||Grade Limit|
Impact of changes on lender returns.
It is forecast that the portfolio realised annual return (RAR) for retail lenders will drop by 0.6% as a result of the net impact of the changes in interest rates and pD rates. This forecast was based on comparing the performance of the entire portfolio of Harmoney loans from January 2015 to May 2017 under existing pricing and default rates with the new interest rates and default rates applying from 3 August 2017. The analysis was based on a combination of current and forecast performance and may not be a good indicator of future returns. The actual change in return as a result of the changes will depend on your lending criteria and what mix of risk grades you lend to, and your applicable Lending Fee.
|Interest Rate p.a.||21.1%||19.1%||-2.0%||Lower borrower interest rates reflect increased scorecard accuracy and a lowering of expected credit default by grade. Borrowers will now pay 10% less interest on their loan on average.|
|Probability of Default p.a.||1.98%||1.63%||-0.34%||The lower forecast losses due to charge off will reduce due to better prediction of loss.|
|Historical RAR impact (after defaults and fees)||12.7%||12.1%||-0.6%||The interest that Lenders receive will be lower. This is offset by lower forecast losses.
Lower Borrower interest rates may also provide further positive changes for Lenders on the Harmoney platform:
Note: The Lender fee used in this example is 17.50% (lender fees are 15.0%, 17.5%, or 20.0% depending on your tier).
- Why do I need to turn my Auto-lend back on?
- You need to do this as the lending parameters are changing (interest rates, loan limits, pD rates) and Auto-lend will be de-activated when this occurs. You should only turn Auto-lend back on if you are comfortable lending under the new parameters.
- Why are Borrower interest rates being reduced?
- The new scorecard has recalibrated the risk grades and the type of Borrower in each grade. This combined with the improved accuracy of the scorecard provides greater certainty that borrowers have been graded correctly and also it means that interest rates can be sharpened. These should result in a higher volume of loans benefitting more borrowers and lenders.
- Is a Borrower that is an C3 in the old scorecard the same as a borrower that is C3 under the new scorecard?
- Scorecard 1.5 has recalibrated the risk grades so a C3 borrower under the old scorecard is likely to be different to a C3 under the new scorecard. This is reflected in the interest pD rate.
- Do these changes apply to institutional lenders as well.
- They apply to all Borrowers and therefore to institutional and retail Lenders alike.
- What are the forecast projections for how the new interest rates will affect the current platform/retail RAR?
- It is forecast that the portfolio RAR for retail lenders will drop by 0.6% as a result of the net impact of the changes in interest rates and pD rates. This analysis was based on comparing the performance of the portfolio of all Harmoney loans from January 2015 to May 2017 under existing pricing and default rates and the new interest rates and pD rates. The actual change in return you will experience as a result of the changes will depend on your lending criteria and what mix of risk grades you lend to and lending fee tier. The analysis was based on historic performance and forecast of future loan performance and may not be a good indicator of future returns.
- How will the changes affect Rewrites?
- All loan re-writes that are applied for on or after 3 August 2017 will be subject to the new interest rates and loan limits.
- If a Borrower adds Payment Protect to their loan – will this have any bearing on their risk grade?
- No, it will not.
- How will this affect my current Auto-Lend settings? (interest rate, est. default rate etc).
- The Interest Rate and default rate filters are being re-calibrated for the new interest rates and default rates applying from 3 August 2017. Before turning Auto-Lend back on you should review your settings and the new interest rates and default rates and only turn Auto-Lend back on if you are comfortable lending under the new parameters.
- Do these changes affect my existing portfolio?
- The changes apply to new loan applications only. Existing loans in your portfolio do not change in any way.