Hello valued visitors! We're bringing back Week in Risk but calling it Week in Financial Education until we find a better title. Maybe you noticed our exciting news? We recently joined CeriFi and look forward to broadening this forum's reach. The idea is to share a brief blog-like update from the week including: selected forum threads that highlight interesting exam-related topics, curated links related to our domain(s), and--if I have something to add--my musings as a passionate student of risk, finance and data science (my practice also informs my equities portfolio construction which I sporadically write about over on Seeking Alpha). Please let me know what you think! You know where to find me ...
In the forum (beginners and new learners)
Upstart ($UPST) emerges as a fintech (credit risk) star. As an investor (and student of risk), my most compelling read last week was Upstart's quarterly earnings call. The market rewarded its performance with a +39% pop. This company ($UPST) is disrupting the credit business with an artificial intelligence (AI) lending platform. By applying data science to alternative (and big) data, they not only double/triple approval rates (while holding losses constant) but they enable banks to extend credit to underserved, perhaps even unbanked, populations. The application of machine learning to underwriting feels like the inexorable future. Said the CEO, "while most lenders consider only a handful of variables as part of a lending decision, Upstart's model considers more than 1,000 variables about each applicant. You can think of these as the columns in a spreadsheet. And as of December 31, 2020, our model was trained on more than 10.5 million unique repayment events." When I took my first machine learning class (in person, a few years ago), upon learning random forests--which are a black box to most of us--the class had a debate about the implications of data science on lending practices. My initial reaction, in all sincerity, was a bit of horror: how can banks make lending decisions based on black boxes algos? (If you think I exaggerate, the winner of our contest literally could not articulate why his algo outperformed). But participants smarter than me argued how algos can be more fair (should I say more equitable?). In any case, in the FRM of course we learn about traditional credit analysis and scoring (e.g., borrower capacity and willingness). And that's good. We need those fundamentals. But the future of credit analysis looks very different and understanding it will require additional skills.
Interest rate (or is it inflation?) risk is the market risk everyone is talking about. Howard Marks of Oaktree writes that the market's biggest risk is the possibility of risking interest rates. Ben Carlson says that inflation matters more than interest rates. Professor Aswath Damodaran explains the relationship between interest rates and inflation. The Atlanta Fed has a cool Underlying Inflation Dashboard at https://trtl.bz/atlanta-fed-inflation-dash. Fisher Investments has this Q&A on inflation. Where do I learn more? One source is the phenomenal Lynn Alden Schwartzer who regularly dazzles the grizzled Seeking Alpha veterans. Here she is on the three types of inflation (monetary, asset price, and consumer goods/services) and here she is on Interest Rate Risk (and the equity risk premium).
In the forum (beginners and new learners)
- Somebody else's CAPM practice question contains an instructive misunderstanding about a key difference between the CML and the SML https://forum.bionicturtle.com/threads/mean-variance-analysis.23736/
- We got two questions about the commodity lease rate which perennially vexes new learners. GARP's new material takes a backward step from previous authors (McDonald remains better) and doesn't attempt numerical reconciliation with the other cost of carry factors. In the basic ("naïve") version, the least rate is quasi-synonymous with the convenience yield; in this naïve version, as both subtract from the forward price (via conferring benefits to commodity ownership) either explains the entire difference between the observed futures price and the constructed cost of carry. In the technical ("sophisticated") version, the lease rate is a net convenience yield. My strong preference is to view the lease rate as the net convenience yield; i.e., L = y - u. This will reconcile dicey applications, in particular when gold has simultaneously both a convenience yield and a lease rate. See https://forum.bionicturtle.com/thre...-carry-theory-hull-chapter-5.10611/post-87786
- Rohit asks if anybody is pursuing the CFA while in university at https://forum.bionicturtle.com/thre...ertificial-chartered-financial-analyst.23747/.
- How much does the FRM want you to know about post-2008 regulatory responses to the global financial crisis (GFC), including Basel and Dodd-Frank? My opinion is that GARP has always asked for too much review of the Basel regulations. Literal compliance with the learning outcomes (LOs) will probably lead you to regulatory overexposure https://forum.bionicturtle.com/thre...responses-and-best-practices.23058/post-87759
- For some reason, we got several questions this week about using the classic Z-lookup table; e.g., https://forum.bionicturtle.com/thre...-deviate-in-spectral-measure.23708/post-87714. GARP's Z-lookup table displays only negative values, Z = {-3.00 to zero}. FRM candidates need to be totally comfortable with distributional lookup tables. These lookup tables are partly how we talk about distributions. Much of risk is applied math, and much of that applied math is distributions. Distributions are how we numerically capture uncertainty. To be comfortable with the lookup table requires two skills. First, we need to be able to apply the the normal distribution's symmetrical truth that N(Z) = 1 - N(-Z) in order to retrieve any of the values on the un-displayed right side. For example, what's the Pr(Z<1.65)? It is given by 1 - N(-1.65) = 1 - 0.0495 ≅ 5%. Second, we need to be able to invert: N(2.33) = 99.0% is an instance of the symbolic N(Z) = p. That's solving for a probability given an quantile. To invert is to solve for a quantile given a probability: N^(-1)(p) = Z, which in this instance is given by N^(-1)(99.0%) = 2.33. Once we get this, it's easy to see how the lookup table is also giving us p-values (when the test statistic is normal). You can't read (or talk) about this to master it, sorry. You have to practice it.
- My thanks to Lu Shu Kai (@lushukai) who helped us greatly with at least three glitches (i.e., imprecisions or outright errors) this week. In particular, I mishandled Box-Pierce test's null hypothesis. Do you know: Say we observe high p-values in the Box-Pierce test. Do high p-values tend to support or disqualify a white noise time series? See https://forum.bionicturtle.com/thre...t-and-model-selection-with-aic-and-bic.23558/. He also suggested a sweet edit to my long/short hedge question at https://forum.bionicturtle.com/threads/p1-t3-710-long-and-short-hedges-hull-chapter-3.10546/. You would think short hedge circumstances are straightforward! Oh my, they are not. The different scenarios that build on the classic short hedge scenario (i.e., commodity farmer who plans to sell in the future) have been plentiful. I really appreciate his edit that improves the quality of our Q&A.
- Eustice would like some help with the intuition around conditional independence https://forum.bionicturtle.com/threads/conditional-indepedence.23742/post-87743. GARP added conditional independence last year.
- The Story So Far (A Recap of Net Interest Themes) https://www.netinterest.co/p/the-story-so-far Marc Rubenstein's brilliant finance/fintech substack ("Net Interest") is a must-read and he just reviewed some of this themes including financial regulation and the new power brokers (can you guess where he thinks the power lies?).
- SPACs account for a whopping ~70% of this year's IPOs to date (almost $100 billion raised in 2021 so far!), says Dealogic. A helpful introduction to SPACs via interview of SPACInsider's founder: A Blank Check is All the Rage: https://www.brunswickgroup.com/spacs-market-database-i18370/ And today's WSJ article impresses with dataviz SPACs Are the Stock Market’s Hottest Trend. Here’s How They Work.
- Marsh's Global Risks Report 2021 at https://www.marsh.com/us/insights/risk-in-context/global-risks-report-2021-priorities.html
- Catastrophe risk management (and modeling) seems like a truly exciting, multidisciplinary career: 4 questions for Hyeji Kang, Head of Reinsurance And Catastrophe Risk Management https://www.agcs.allianz.com/news-and-insights/expert-risk-articles/grd-cat-risk-management.html
- Norman Marks' Two great pieces on cybersecurity and business risk
Upstart ($UPST) emerges as a fintech (credit risk) star. As an investor (and student of risk), my most compelling read last week was Upstart's quarterly earnings call. The market rewarded its performance with a +39% pop. This company ($UPST) is disrupting the credit business with an artificial intelligence (AI) lending platform. By applying data science to alternative (and big) data, they not only double/triple approval rates (while holding losses constant) but they enable banks to extend credit to underserved, perhaps even unbanked, populations. The application of machine learning to underwriting feels like the inexorable future. Said the CEO, "while most lenders consider only a handful of variables as part of a lending decision, Upstart's model considers more than 1,000 variables about each applicant. You can think of these as the columns in a spreadsheet. And as of December 31, 2020, our model was trained on more than 10.5 million unique repayment events." When I took my first machine learning class (in person, a few years ago), upon learning random forests--which are a black box to most of us--the class had a debate about the implications of data science on lending practices. My initial reaction, in all sincerity, was a bit of horror: how can banks make lending decisions based on black boxes algos? (If you think I exaggerate, the winner of our contest literally could not articulate why his algo outperformed). But participants smarter than me argued how algos can be more fair (should I say more equitable?). In any case, in the FRM of course we learn about traditional credit analysis and scoring (e.g., borrower capacity and willingness). And that's good. We need those fundamentals. But the future of credit analysis looks very different and understanding it will require additional skills.
Interest rate (or is it inflation?) risk is the market risk everyone is talking about. Howard Marks of Oaktree writes that the market's biggest risk is the possibility of risking interest rates. Ben Carlson says that inflation matters more than interest rates. Professor Aswath Damodaran explains the relationship between interest rates and inflation. The Atlanta Fed has a cool Underlying Inflation Dashboard at https://trtl.bz/atlanta-fed-inflation-dash. Fisher Investments has this Q&A on inflation. Where do I learn more? One source is the phenomenal Lynn Alden Schwartzer who regularly dazzles the grizzled Seeking Alpha veterans. Here she is on the three types of inflation (monetary, asset price, and consumer goods/services) and here she is on Interest Rate Risk (and the equity risk premium).
Last edited by a moderator:
.