CUES Member Community

Expand all | Collapse all

Cecil and Loan Analytics solution

  • 1.  Cecil and Loan Analytics solution

    Posted Nov 25, 2020 12:19 PM
    Just wondering if anyone out in CUES land would be willing to share their experience with vendors for a Cecil and Loan Analytics solution.  Who are you using? How do you like it?  For Cecil, anyone doing parallel runs?

    Thank you in advance for those that are willing to share.

    Brad

    ------------------------------
    Brad Hudson CPS CSE
    Director of Consumer Lending
    University of Kentucky FCU
    Lexington KY
    859.257.2678
    ------------------------------


  • 2.  RE: Cecil and Loan Analytics solution

    Posted Nov 27, 2020 05:49 AM
    We built ours in house.  We use loan life cycle data for the models developed in R.  I spoke with a couple different vendors and if you are fine with a vanilla model fr predictions that is used for every credit union than they are all doing the same thing.  If you want to be able to customize the model, or if you have some unique challenges other credit unions might not have (i.e. large participation portfolios) than you would either want to develop your own or have some conversations with them regarding how to account for those in the model you are purchasing.

    ------------------------------
    Troy Del Valle
    VP, Business Intelligence
    Hudson Valley Credit Union
    Poughkeepsie NY
    845.463.3011
    ------------------------------



  • 3.  RE: Cecil and Loan Analytics solution

    Posted Nov 27, 2020 09:39 AM
    Brad,
    You mentioned two issues (loan analytics; CECL) which are highly related and yet can have separate solutions, and the right selection for a CU likely varies with the need and expected use, along with asset size and complexity/characteristics of the loan portfolio.

    Our situation: We are $1.15 billion in assets, and have used a multi-dimensional loan analytics platform (Lending Insights) for close to 10 years, far in advance of CECL. We have been heavy, consistent users of that platform and use it for static pool analysis, LTV analytics, credit tier analytics, indirect dealer profitability, and more. Whether a CU already has an adequate loan analytics solution, and/or whether a CU is likely to actually utilize what the analytics tell them for managing the loan portfolio (such as pricing or underwriting changes) is a consideration in selecting an analytics and/or CECL vendor.

    We have also used an Allowance model (ARCSys) for about 8 years for the incurred loss calculations, and initially expected (and started down the path) to migrate to their CECL model. I've also seen multiple demos from quality companies such as 20/20 Analytics, Sageworks (now Abrigo), and Visible Equity. There are many other vendors out there as well.

    The single biggest problem for virtually every bank and CU for implementing CECL is "data." FASB wrote a standard which technically can require monthly loan/instrument-level data for every loan for 10 to 12 years lookback, when few institutions had accumulated data at that level of granularity or consistency. Lenders are "supposed" to have this data for the contractual life of the loans in the portfolio, not just for the average life. When we talked with our current vendor about migrating to their CECL model, the discussion quickly resolved to "Well, we see that you have great data back to January 2013, but what about the previous 4 or 5 years?"  "Also, what about data on the loans you acquired in a couple of mergers?" We did a core conversion at the end of 2012, and simply have/had no way to recreate data to satisfy CECL. Our vendor began suggesting that we could recreate data from AIRES files or our ProfitStar ALM files, but we had NEVER had a need to retain those types of files for that length of time. Then, a suggestion that we could engage a company to read monthly microfiche for those earlier years and create an loan/instrument-level "data shell."  Of course, that would cost tens of thousands of $ and still provide very incomplete data, and would probably take well into 2021 just to get data created.

    If a CU needs and will use analytics data, it may still be a good decision to utilize one of the high-level, complex platforms such as 20/20 or Visible Equity, etc.  You'll need to have up-front discussions on exactly how the CU and the vendor will address issues for data availability and consistency and the length of lookback; that discussion will be influenced by the nature of the CU's loan portfolio in terms of complexity and mix. These types of systems can (or at least, look for a solution that's not "just" a CECL solution) provide valuable analytics beyond the CECL calculation.

    Our change in direction: We were contacted by our audit firm (Clifton Larson Allen) about considering a WARM CECL model. Now I'll tell you (from personal experience) that the vendors who are out there selling solutions with maximum complexification (my made-up word) will rake you over the coals about the evils and inadequacies of a WARM solution. In our case, we actually used to use a WARM-oriented spreadsheet calculation which took annualized loss ratios, incorporated QE factors, and we then multiplied by the average life of the pool. I always felt that there were more than 1 year of losses in a loan portfolio, but we were told by our auditors to stop doing that  because we were "overstating" our ALLL. I guess FASB finally reached the same conclusion. We were pretty comfortable with using a WARM methodology and incorporating QE factors, etc., so we had a detailed discussion over specific questions about WARM acceptability for CECL. Further, I went through the records & reports of the 2018 and 2019 joint FFIEC and FASB conferences about the acceptability of WARM. Other CUs considering WARM should also review these, but generally the more uniform and less complex a lender's portfolio is, the more WARM can be a good solution for CECL. We explicitly asked our auditor for their interpretation of the "mix and complexity" points, as well as about outright size of a loan portfolio.  We have chosen to use the CU-Metrics WARM model. It automatically loads 44 quarters of Call Report data, and starts with some default assumptions and calculations about loss history and average life of the pools. The pools are at the Call Report level, though if more granularity is needed, side calculations can be made using spreadsheets and then loaded into the model. CU-Metrics incorporates structured QE adjustments and structured future loss estimations. You can readily see how the CECL calculation relates to your current ALLL balance. We are working on ours, and will be running parallel no later than Q1-2021.

    We are very comfortable with this approach, but we are also working to fill in data gaps as we move forward, so if future circumstances require us to migrate to a more-sophisticated model, we will presumably have the needed data at that time. If we did not have our Lending Insights platform for loan analytics, I may have decided differently in order to have a combined analytics/CECL solution. I think CUs have been "sold" on the need for complex modeling solutions by the vendors selling those complex models, when many CUs are unlikely to achieve a materially "better" ESTIMATE for the Allowance by using complex methods over simpler WARM methodology. Personally, I think complex CECL models provide a false perception of precision and lead us to forget that it is, and will always remain, an ESTIMATE. WARM is unlikely to pick up emerging shifts in underwriting, etc., so there are trade-offs. Discuss with your audit firm in making your decision.

    Two final points: 1) Don't overlook WARM as a potential CECL solution, but consider the CU's variability of losses and complexity of the loans in your portfolio, and what additional analytics value the CU might utilize from a more-complex solution. 2) You absolutely must get started ASAP, whatever CECL solution you select. I don't believe there will be any further delays, so there are only 2 years left before the January 2023 implementation. The sooner you can begin to deal with data issues, and to get initial estimates of the impact on the Allowance, the better positioned you will be.

    There's no one best solution, so good luck to us all...​

    ------------------------------
    Paul Meissner CCUE, MBA
    Chief Financial Officer
    Credit Union of America
    Wichita KS
    316.265.3272
    ------------------------------