“I don’t need to know how the watch works. Just tell me what time it is.”
- Millennium PM, circa 2007
That quote comes from a colleague of mine roughly 13 years ago when I was on the buy side. I overheard him saying it often when doing calls with sell-side analysts who would ramble on about details he didn’t care about or think mattered. More than a decade on, I’m reminded of that quote whenever I’m asked how our new AI [Artificial Intelligence] Stock Screener works, but before we get into that, first, some context.
Pushing back roughly a decade earlier, the year was 1996, and after 7 years of research and development a team at IBM believed they were ready to challenge Garry Kasparov, the reigning World Champion in chess. In game 1 of 6 Deep Blue eked out its first win against the Champion, but Kasparov learned fast and won the match decisively.
The following year, Kasparov confidently proposed a rematch which was immediately accepted by IBM, for they had been learning too. On the afternoon of May 11th, 1997 – game 6 – the deciding game, Deep Blue’s 256 processors enabled it to consider 200 million positions per second; whereas it’s been estimated that Kasparov could consider just two. Deep Blue went on to beat Kasparov in 19 moves and won the match – the first time a computer had ever beaten the World Champion.
|
The Deep Blue victory has since entered into the cultural lexicon as a major milestone in the advancement of AI, but the distance of time and the benefit of hindsight makes it easy to overlook the fact that, when Kasparov sat down to play, he firmly expected to win. “I don’t think it’s an appropriate thing to discuss the situation if I lose,” Kasparov said leading up to the match. “I never lost in my life.” When Kasparov then lost, and lost in dispiriting fashion—in Game 2, he described the computer as playing “like a god for one moment”—he seemed to have been not only intellectually but spiritually defeated. From The New York Times’ coverage of the match: “‘I was not in the mood of playing at all,’ he said, adding that after Game 5 on Saturday, he had become so dispirited that he felt the match was already over. Asked why, he said: ‘I’m a human being. When I see something that is well beyond my understanding, I’m afraid.’” – Kevin Lincoln, theringer.com |
While Deep Blue was retired by IBM after that match, the world of chess computers continued to evolve and advance. ELO ratings are used to gauge abilities of players. Still today, Garry Kasparov has the second highest ELO rating ever recorded for a human at 2,851, surpassed only by current World Champion Magnus Carlsen at 2,882. An ELO rating advantage of 100 points equates to a 64% probability of winning, 200 points = 75.8%, 300 points = 85.3%.
The current top ranked computer chess program is Stockfish and it has an ELO rating of 3,496, a 614 point advantage over the current human World Champion. That equates to a 98.2% probability of winning. Meanwhile, the edge between man vs machine continues to widen. It’s been 23 years since Deep Blue’s victory, but the top human’s ELO rating has advanced just 31 points. Chess computer’s ELO ratings, meanwhile, have advanced by 645 points over that same time period.
For those who would argue the stock market is a far more complex system than a chess board, consider this. According to RankRed, there are 10120 possible moves in chess. To put that in perspective, there have been 1026 nanoseconds since the Big Bang and there are an estimated 1075 atoms in the entire universe. Chess is, indeed, a complex system.
This intro naturally brings us to our latest product innovation: our Financials and Housing AI stock screeners, which we began working on just over two years ago. At the risk of gross oversimplification, our approach was to develop a 200-factor nonlinear, nonparametric machine learning system that would treat the market as a classification problem and attempt to capture the non-linear effects that are found in the stock market. The goal was to minimize training loss function (i.e. find the best fit) and penalize overfitting. This strategy not only works better for nonlinear data, but also enables lower computational intensity allowing us to process and train on far more data.
In plain English, we built a chess computer for ~350 Financial and Housing stocks, and then we produced and evaluated out-of-sample results over several months to make sure it worked. The results were impressive. The chart at the end of this note shows the 12-month performance of the longs and shorts for each weekly iteration that we tested out-of-sample. Going forward, we will publish the AI Stock Screener’s top ideas, both long and short, every week on Monday morning for our Institutional and Financials Sector Pro subscribers. We recently presented a detailed overview of our system, process and results. Let us know if you’d like to know more.
Back to the Global Macro Grind ….
Consumer lending companies tend to be very cyclical. This owes mainly to their credit sensitivity. In normal times, when unemployment is low, charge-off rates on unsecured consumer receivables (credit card loans) tend to run in the 3-5% range. During periods of stress, loan losses can rise significantly. At the peak of the Financial Crisis, annualized charge-off rates reached 10.6%. Looking back over the last 20 years, the historical relationship has been that for every 1 million increase in continuing weekly jobless claims, industry credit card charge-off rates have risen by 1.7%.
This is important because, as of last week, continuing state unemployment insurance claims still stood at 16.2 million – 4 months into the crisis. That number alone would equate to an annualized charge-off rate of around 27%. Beyond this, there are many millions more unemployed workers collecting PUA (Pandemic Unemployment Insurance) benefits under the CARES Act. Strikingly, charge-off rates for credit cards sit at just 3.6% as of the end of Q1 with little movement, in fact some improvement, based on the numbers released so far for Q2. How can this be the case and what are the go-forward implications?
To begin, let’s consider last week’s comments from Rich Fairbank, CEO of Capital One:
|
The huge sort of elephant in the room on the consumer side and it's an elephant on the commercial side as well is what happens to government stimulus. And I'd just think a lot of things have lined up that have softened the impact for consumers, even really those who have been unemployed. And so, we are seeing this great paradox of extraordinary credit performance in the middle of the worst economy metrics in our lifetimes. So I think that's a hard one to prognosticate where it goes from here. |
The transfer payment [government stimulus] component is enormous. A key player is the enhanced unemployment benefit authorized under the CARES Act that allowed for a supplemental payment of $600/week in addition to the baseline state benefit regardless of previous income or geography. The average national state unemployment benefit is $385/week. Coupled with the $600 that works out to >$50k/year – enough to make it so that 60% of benefit recipients are currently earning more from unemployment than from their previous job. The takeaway is that it is this supplemental payment that has enabled, in many cases, those who have lost their job due to Covid to keep current on their mortgage, rent, auto loans, student loans and credit cards. Moreover, lenders have aggressively offered forbearance programs that have helped give borrowers additional breathing room.
This brings us to Washington. Senate Majority Leader McConnell’s initial salvo is to lower that $600/week supplemental payment to $200/week. This would reduce earnings from $985/week to $585/week – an effective 40% ‘pay’ cut for well over 16 million people. What’s more, the plan is for this to last just 2 months before dropping down to 70% of the recipient’s previous income. This assumes there are enough Cobol Cowboys left to reprogram State unemployment software to enable this provision. Let’s consider a few examples. According to the BLS, here are the average weekly earnings for
- Food Service & Drinking Places workers: $390/wk
- $390/wk → $985/wk (+150%) → $273 (-70%)
- Leisure & Hospitality: $427/wk
- Child Care Services: $482/wk
- Arts, Entertainment & Recreation: $537/wk
- Retail Workers: $605/wk
This means that Food Service workers, who saw their income rise +150% from $390/wk to $985/wk, would see their weekly income fall from $985 to $273 (70% of $390) – a 72% reduction – two months from now. Hotel workers income would fall by 70% and so on. This isn’t intended to be a political statement. Objectively, these are massive, looming income hits to people who largely have little to nothing saved.
The risk embedded in this 2H20 dynamic is why we continue to see outsized risk in the intermediate term for card lenders SYF, COF, DFS and AXP.
On the other side of the trade, however, are the debt collectors, ECPG and PRAA. These companies benefit from rising levels of loan losses because they are the third-party buyers of charged-off paper. More supply tends to coincide with lower demand and that means better pricing for the buyer. Better pricing = higher earnings. It is not coincidence that the debt collectors have consistently been the top-performing subsector of Financials in the early stage of every economic recovery, rising 10-fold coming out of both of the last two recessions.
Incidentally, we’re not the only ones who like ECPG. Our AI Stock Screener also likes it.
Immediate-term @Hedgeye Risk Range with TREND signal in brackets:
UST 10yr Yield 0.57-0.64% (bearish)
SPX 3175-3277 (bullish)
RUT 1 (bearish)
NASDAQ 10,266-10,812 (bullish)
Tech (XLK) 104.21-110.16 (bullish)
REITS (XLRE) 34.32-36.36 (bullish)
Utilities (XLU) 58.37-61.36 (bullish)
Financials (XLF) 22.81-24.55 (bearish)
Shanghai Comp 3150-3384 (bullish)
Nikkei 226 (neutral)
VIX 23.26-30.88 (bearish)
USD 93.02-94.99 (bearish)
Oil (WTI) 39.93-42.19 (bullish)
Nat Gas 1.60-1.93 (bullish)
Gold 1 (bullish)
To your continued success,
Josh Steiner
Managing Director
