The big puzzle of EU banking is how much loan moratorium will translate into loan losses. Nationwide has useful disclosure on this, because they are the only (afaik) ones making a clear distinction between moratorium now expired or not and what happened to expired moratorium.
So what does it say?
On prime mortgages, only the moratoriums from March and April have expired, that's 99k mortgages. 86k of those have now resumed normal payments. Which means there are approx 13% of "deliquencies"
But there are different types of problems.
On 8k mortgages, they have so-called partial payment plans agreed, which means foreborne exposure in the banking jargon. It's not good, but it's not bad as a default. For 5k exposures, they sought "further support", which is a nice way of saying borrowers can't pay.
Bank of Ireland (yes, I'm very Irish today) has some disclosure that illustrates the massive shock of Covid on the lending market, with the movements in their (internal) ratings. It's quite stunning 1/n
Here's what it looks like on the entire book.
Blue is investment grade, orange is BB, so medium risk, grey is high risk and yellow is real shit.
Down from 46% of the book being considered (super) safe, to only 7% !!!
From 7% shit to... 17% shit ! This is a massive shift.
And I suspect Bank of Ireland is not worse than other countries which are lending on a LTV basis (some countries are more lending on income basis), they are just more transparent.
Some book by book disclosure shows what's happening.
A thread inspired by two charts (h/t @C_Barraud & @economistmeg ) which raise an important question: what will happen to banks if there’s a 2nd wave or 2nd lockdown?
First, an observation: I believe cases will soon get to the all time peaks we saw during lockdowns, but the actual health situation is very different for three reasons: i) we’re better at curing sick people ii) infected people are younger and more importantly...
iii) the real number of cases back in March/April was probably an order of magnitude higher - but we didn’t really do tests so we didn’t know.
A few thoughts on the A level debacle because I think it shows very important trend - especially when compared to what happened in France, which obviously faced a similar situation. How did both country do?
This is quite stunning.
Basically the UK tried to use a very sophisticated algorithm with rather fancy maths - the purpose was to be fair, not only for bad grades but also good ones (I'll set aside claims that all the problems were actually intended- I don't believe government is smart enough for that)
Of course it was never going to work for all, of course there were going to be problems , errors, but it was an impossible task - no one can reproduce what the exam would do without actually doing the exam! Algos are not wizards
I wanted to share a crazy story.
Asset managers have to report risk ratios and to stay below limits. Sometimes, big 4 auditors are appointed to calculate those ratios.
One of them (no names here!) is using its "own" model to compute the risk you're taking on AT1/Coco bonds. 1/n
I got puzzled when the report showed that I had two bonds (one rated BB+ the other BB), one with a risk of 4% and the other with a risk of 40% (just so that you know: for me the risk is 100%!!)
The numbers did not make any sense. So I investigated. Turns out this big 4 auditor (for all its clients worldwide) is relying on some model designed ages ago by someone who wasn't even graduated from college. I kid you not.
Every single client they have is "using" that model
Maybe it's because lots of paper aren't peer reviewed, maybe it's because diff fields diff expectations, but there is something deeply disturbing about scientific research on Covid.
Most papers that use statistical methods are deeply flawed and not even in a sophisticated way
I am interested by Twitter discussions on masks, lockdowns, airborne transmission, HCQ and they often end by someone tweeting a link to some publication - supposedly the final argument.
But depressingly it very often takes less than 5 minutes to find a major flaw
A few examples.
A piece on the impact of the french local elections on Covid in France was widely publicized. As @Emmanuel_LEVY wrote, the stats are total nonsense.
A few thoughts on the ECB Financial Stability Review and their analysis on banks' asset quality (while we wait for the "desktop" stress test.) There are quite a few interesting dta points in there.
Very much like the BOE, the ECB wants to analyze the cash flows weaknesses of sectors impacted by Covid and see how it translates into loan losses. So the starting point is this chart showing the share of such sectors in total VA
Clearly, it's a massive share of the economy.
then they look at refinancing profiles of sectors, in this chart.
It's not exactly lucky that one of the worst sector (hotels and restaurants) has the worst refi profile... but on absolute numbers, the biggest risk is retail (though you would have to remove food from that imo)
HUGE: I have done a worldwide analysis to show the causal link between lockdown policies and the reproduction number of the pandemic. This is of crucial importance to tell if the lockdowns were medically (not to mention morally) justified.
I am now ready to share my conclusions.
And the conclusion is:
I HAVE NO F***** CLUE.
And I really spent a lot of time trying to use all sorts of causality tests, simulations, trend change tests, ML classification, etc.
Also I have read articles claiming that LD reduced R and some claiming LD didn't reduce R. And I really believe their robustness is zilch. All of them.
So here you are : you can continue enjoying your priors ! Isn't life wonderful?
Let's continue the digging on this. Another important fact we have to bear in mind is that lockdown is not digital. It's not ON/OFF. It's a continuum of measures (social distancing, school closures, social events, etc.) Same, relax of lockdown is not ON/OFF, it's also gradual.
So if want to see how it changes R, we need to quantify the lockdowns. Luckily, Oxford econ has you covered as they have a team of 100 monitoring lockdowns over the world and creating a "stringency index" which quantifies the measures taken to reduce contacts (so reduce R).
Rather than some absurd R change over 1 week, let's look graphically (often better than shitty statistical analysis) at how R and "Lockdown Stringency index" moved over time!
I don’t min shitty statistical analysis, I read some every day. But the problem is when it comes with pompous and grand statements such as “In the absence of conclusive data, these lockdowns were justified initially.” But “millions of lives were being destroyed .. "
"with little consideration that [lockdowns] might not only cause economic devastation but potentially more deaths than COVID-19 itself.""
So I’ve spent half the afternoon looking at Greek banks in a Covid world and now I want to drown myself in the Aegean sea. A thread.
How did they enter the crisis? Well capitalized, you might think. CET1 ratios between 14.9% (Alpha) and 12% (Piraeus), wow that sounds great. But there’s a catch. A BIG chunk of that is deferred tax credit.
What’s that? I won’t go into the details, it’s money the government is supposed to “repay” if things go south, but so far it never actually worked as hoped. Monkey money if you ask me. But for some banks it’s almost 100% of the CET1!
This is v. interesting - basically oxford econ did a huge database of world lockdown measures and ONS regressed it against known GDP prints. It works pretty well, so I applied it to latest values of "lockdown index". Here's the current cumulated Q1+Q2 estimates for some ctrys
(Countries were chosen mostly because they're relevant for banking risk - sorry that's my bias!)
Italy, France and Spain not really a surprise. More surprised by China, Turkey or even Russia.
Obviously, apart from China/SK the regression was mostly done on western Europe/ US/Can so not sure it works for EM countries. It will be worth updating this as we get more prints (and maybe improve regression with other variables)
Why did I make that promise? I must be mad! Anyway, here it is, The Deutsche Covid stress – and seriously, you’d better switch immediately to your favourite Trump or Covid meme threads, because this could be long and geeky! 1/N
How do we approach this? The goal is to see what their Common Equity Tier 1 (CET1) ratio could look like by FY2020. Maybe you think CET1 is a bad measure, but it’s what supervisors use & they call the shots – so it’s what matters 😊
CET1 ratio is CET1/RWA. Let’s start with CET1. Now that Basel 2 transition is almost over, FY2020 CET1 will mostly be driven by P&L. DB’s P&L is a gigantic mess, with many moving parts, especially as they‘re in a restructuring process.
I have been bugging you with #IFRS 9 for banks and the #Covid crisis, and by now you probably think I’m senile, repeating the same things all over again. ENOUGH WITH THE GEEKY STUFF NO ONE CARES ABOUT.
An (important) thread.
But wait, the Fed, the BoE & the ECB have all published on this (or the US equivalent). And they are all trying to find a way to tweak a rule which is not really in their remit (it’s an accounting rule.) So maybe there *is* actually something important?
I’ll explain why first with the theory and then a STUNNING example. Bear with me through some accounting stuff, before we get to the juicy numbers.
Excellent news on the #Covid front today, my estimates of the reproduction factor R is going down sharply almost everywhere. I believe that this is due to new testing patterns "catching up" in the data so R estimates getting closer to the real values. Here's a sample of countries
Significant drop in the UK, even if remains above 1
The Swiss bank regulator (FINMA) does statistics like no one else. A quick story - to file under "Fat Tails Twitter".
Banks have to hold capital for market risk. This is calculated using sophisticated models that estimate the VaR or the eVaR, VaR is the worst loss you are statistically likely to take with a (say) 99% confidence, eVaR the average loss above that VaR
To check if the models work, regulators (and banks) do back-testing: if you calculate you should loose less than 1m 99% of the time, but actually loose 5m 10% of the time, you have a problem!
I am investigating which banks are sensitive to the #Covid crisis & I thought I’d share a very specific situation as it illustrates how tricky it is to analyse a bank’s balance sheet and risks – and the weirdness of some capital regulations.
Let’s talk about Barclays. (You know the rules: this is NOT investment advice.) People always say "banks’ internal models are a black box and are “gamed”", so let’s look at Barclay’s book which is NOT using internal models.
For Barclays, that’s a 270bn£ book, in terms of exposures at default : the metric you multiply by the “risk weight”, which measures the risk of the exposure, to get the Risk Weighted Assets (RWA) ; capital required is approximately 12% of RWA
So let me get this hilarious story straight. NATIXIS.
This bank has a big oil exposure, owns 51% of H2O which sees its funds crash and ofc is Covid exposed like the rest of the world. unsurprisingly the stock crashes.
But wait, what happens mid march ????
Insane rally and epic short squeeze ! Why ? Because of the crash the dividend yield goes to the roof ! And the bank replaces Unibail in a high dividend index, which triggers massive forced... buyers ! (ETF, index funds etc)
But with a cooler head, I'd like to point up that the package is a disappointment on govies/rates, but it's MASSIVE for banks.
Just the new TLTRO thing is huge: the loan growth benchmark is now 0 (LOL), the amount available is now €2.3tn - and this will be at -0.75% And if you don't know what the do with the cash, depo at -0.5%. U can do the math of this subsidy.
And what about this buffer thing the SSM announced? Pretty sure the algos didn't pick up on that one. But if you add 2.5% (CCB) + 1.5% (average P2G) and multiply by the approx. 20tn€ RWA in the European banking sector...