Hacker News Reader

Comment by hn_throwaway_99 | Hacker News Reader

> Is it because some banks still depend on mainframes doing batch processing?

Pretty much all banks use mainframes to do batch processing. "SFTP runs the world".

I got heavy into fintech a couple years ago, and at first I thought "This is all insane." I was coming from a SaaS world where it's easy to think about transactional guarantees with API calls.

I finally "got it" when I realized that pretty much all of our financial system in the US is based on the digitization of manual processes from the 50s and 60s. If you think of how banks worked before computers (e.g. large binders of ledgers that were manually reconciled), that's basically how it still works today, those big binders have just been turned into digital files that are processed by machines.

I'm no fan of crypto, but the one place where I can see real utility for blockchain is as a backend settlement service for banks.

View on HN

> Pretty much all banks use mainframes to do batch processing. "SFTP runs the world"

This isn't an anachronism. Batched net settlement is inherently cheaper than real-time gross settlement.

Canonical example: if I send you $10, then you send me $5 and I send you $2, in net settlement, there is a single $7 transaction; with RTGS there are three transactions of $17. The first system not only has lower transaction costs, it also scales better on account of needing less capital. TL; DR If there is a real-time settlement system, it is always possible to build a net settlement system on top of it that's cheaper (albeit slower).

Why should transaction "cost" anything? It is just running arithmetic instruction on a number in a computer. I would argue batch processing makes no sense. If you send me me $10 and I send you $5 and bank wait for a month to run that batch, you might have already been bankrupt and have nothing to send so somehow you sent me money, I never received it and now everything is in big limbo. Transactions should be instant.

> Why should transaction "cost" anything? It is just running arithmetic instruction on a number in a computer.

Every dollar that crosses a payments platform has a non-zero chance of creating some combination of legal, regulatory and customer-service costs. Frictionless transactions are as real world as their physics counterpart.

> would argue batch processing makes no sense

It does, from a deeply fundamental level. The only case where RTGS is equal to net settlement is if we make some Econ 101 assumptions (frictionless transactions, infinite borrowing and lending capacity) together with zero or negative rates. (Even then, one could be a stickler about the cost of computation.)

> you might have already been bankrupt

Clearinghouses.

My point isn't that net settlement or RTGS are inherently better. They're not. But they're fundamentally tied, and net settlement on top of an RTGS rail will always be cheaper in the real world than that same underlying rail. On the cost side, net settlement will win. On immediacy and counterparty risk, RTGS. A strong system offers both options.

> Every dollar that crosses a payments platform has a non-zero chance of creating some combination of legal, regulatory and customer-service costs. Frictionless transactions are as real world as their physics counterpart.

Yes, but pretty much all of those costs are a factor of the originating individual transactions. In reality net settlement is just one additional transaction where things can go wrong.

Again, I'm no fan of crypto (WRT how it's currently used, scams, etc), but there is no reason a permissioned blockchain couldn't do millions of transactions a second, all with real-time settlement, at a cost lower than batch processing.

> all of those costs are a factor of the originating individual transactions

Correct, to a degree. Hub-and-spoke systems (clearinghouse, with banks at the periphery) are cheaper and more efficient than an everything-connected topology. The clearinghouse can run cheap rails because it knows it has pushed those problems to the periphery. The main missing factor, however, is float.

Float makes RTGS more expensive than batched settlement. If you give me RTGS rails, I can batch on top of it and recycle the yield on float into savings, possibly rebates (or credit, e.g. how banks front credit against deposited cheques).

The batch mechanism will always be cheaper under real-world conditions. That's what I want to emphasize. This isn't an artefact. It's fundamentally inescapable.

Net settlement is a useful abstraction. Instead of a fully meshed network with every bank account in the country that was active that day having to be settled against each other, you can reduce the number of nodes to “just” the 14,000 banks and credit unions in the USA. Even that is simplified by running interbank settlement through a central clearinghouse at the Federal Reserve.

As long as operating a connection between two accounts has a non-zero cost and resolving consistency problems has a non-zero cost net settlement will be cheaper and easier.

Tangent: this faster payment layer with delayed settlement is exactly how the Lightning Network on top of Bitcoin works.

>I'm no fan of crypto, but the one place where I can see real utility for blockchain is as a backend settlement service for banks.

So... what does the crypto provide there that a simple authenticated communication protocol with a few redundant centralized servers doesn't?

There are large benefits to not relying on a few redundant centralized servers in that case. We have "clearing houses" now, and in most cases they shouldn't be necessary. If you had something like a "permissioned blockchain", any financial institution could easily validate that everything was settled, in real time.

Well someone has to write the code and run the nodes. I’m not on the side of clearing houses but I don’t see the improvement bid switching to a system run by a crypto foundation and a few blockchain validator companies.

A blockchain will allow all these operators to transact safely. It's really just more secure because it's open (otherwise very quickly compromised). And blockchain tech has shown to be reliable when it comes to up-time and security (if you keep it simple).

> blockchain tech has shown to be reliable when it comes to security

Seriously? You would trust billions of dollars to this?

https://web3isgoinggreat.com/?theme=hack

To be clear, the cryptographic underpinnings of blockchain seem solid, it's just that both implementations and actual usage are by fallible humans. And because everything is irreversible, even if you manage to guard against hackers, you still can straight up incinerate money by (say) mistyping a receiving account number (wallet).

See no mention of Bitcoin or Ether in that page, only centralized shitcoins such as Solana and a few buggy smart contracts. Also, if the system is to be operated by only the banks then transactions _can_ be reversed when needed due to security issues. I don't have a strong opinion whether this is a good option or not, just saying that it's not as absurd as it may seen.

But they are talking about permissioned blockchain which are essentially clearinghouses.

Indeed; check settlement used to be small planes flying the checks to the nearest Federal Reserve every night where all the checks from all member banks would be reconciled and money transferred from the bank where the check was drawn to the bank where it was deposited. Then the checks themselves, having been stamped, would be flown to the originating bank and kept on file. A "bad" check would get stamped and returned to the depositing bank.

The bank where you deposited a check would in fact be lending you the amount of the check until the settlement was complete - hence the daily deposit limits and such to limit the bank's exposure.

Is this still universally the case?

The Eurozone has been working on SEPA Instant Credit Transfer for a few years now. It allows transferring of money within 10 seconds, even to a different bank in a different country. The maximum time is 20 seconds in exceptional circumstances. How would this work with batch processing?

Reply on news.ycombinator.com