Who wants another thread about the absurdity of risk management in the financial sector?

(The crowd goes wild, it’s #RiskManagementMonth #Archegos)

There is a crucial concept in risk management: a group of companies.

Because they are strongly connected, the risk of those companies is similar and highly correlated.

So you want to know if companies belong to the same group.

This is precisely the crucial topic that explains (in part) the bankruptcy of Greensill and the criminal proceedings against its German bank @BondHack & @cynthiao have been talking about:

Was the “Friends of Gupta” a group or not?

This concept is used e.g. by banks to calculate their large exposures limits, by investment funds to calculate their concentration ratios, by insurance companies to calculate their capital ratios, etc.

It's everywhere, the 1st pillar of risk management, who is your risk on?

But how do they do it? How do you know if a company belongs to a group?

Basically, you need (as always) a dataset for reference.

If you’re an investment fund and know your risks well, you can use e.g. Bloomberg and make manual corrections.

But what if you’re a giant organisation? With tens of thousands of clients or more?

You often use a third-party data provider.

Back to little me.

My clients often ask me to produce all sorts of reportings that include that information on a line by line basis.

It’s time consuming, so I decided that maybe I could externalise the work a bit…

So I consulted three providers.

My God what have I done. The can of worms.

Part of the “test” for those providers (all big names) was a bond issued by Banque Fédérative du Crédit Mutuel.

I know, it’s not the easiest one, but it’s a top 3 bank in France, so not exactly small fish.

And it has dozens of billions of bonds issued in the market.

Believe it or not, I got THREE different groups for that same bond.

Three different entities.

Three different legal identifiers.

Maybe you think it’s no big deal, but think about it for a minute.

A large insurance company will have funds invested with many fund managers; each will produce the same kind of reporting…

Now suppose the insurance company has in total 600m€ invested in BFCM bonds…

If the managers have different data providers, the insurance company will never see it has 600M€ on the same group.

It will believe it has 200m€ on three different groups.

A well diversified book.

Except it's not!

And that's just one of the many errors you'll find in those datasets... Good luck regulating that!