What a Goldman Sachs Rates Desk Taught Me About Mortgage Pricing
Helly Shah, former Goldman Sachs interest-rate quant and co-founder & CTO of Ralo, on how mortgage rates are actually made — and why borrowers see so little of that machinery.
Before I co-founded Ralo, I spent my early career at Goldman Sachs in Rates Systematic Market Making, building pricing models and algorithms for interest-rate products. On a rates desk, price discovery is the entire game: everything is quoted, spread, hedged, and re-quoted continuously, and the machinery that produces a price is understood in basis points by everyone in the room.
Then I became a licensed mortgage professional and watched how the same underlying market reaches an actual homebuyer. The contrast is the reason Ralo exists.
Where does a mortgage rate actually come from?
A 30-year mortgage rate is not invented by your lender. It starts from the same global interest-rate complex I worked on — Treasury yields, swap curves, and the mortgage-backed securities market where most loans are ultimately sold. Lenders take that market price and add spreads: for credit risk, for servicing, for hedging pipeline risk, and — the part that varies wildly — for their own cost structure and margin.
Why do two lenders quote the same borrower different rates?
That last spread is where borrowers lose money without knowing it. Two lenders quoting the same borrower on the same day can differ meaningfully, not because either mispriced the bond market, but because one carries a heavier cost structure or simply believes this borrower won’t shop. On a trading floor, a spread like that gets arbitraged away in seconds. In mortgages, it persists for decades, because comparing offers is deliberately tedious and most people give up after one or two quotes.
Shopping is a systematic strategy, not a chore
The quant translation of “shop around for your mortgage” is: sample more quotes from a distribution with real dispersion, and capture the spread between the median and the best. That is not advice, it is an algorithm — which means software should run it, at scale, for every borrower, rather than each family manually re-implementing it during the most stressful purchase of their lives.
That is the core of what we automated at Ralo. The system does what a systematic desk does: it gathers quotes across lenders, normalizes them so they are actually comparable (rate, points, credits, and fees — never just the headline rate), and keeps pressure on the spread until closing. The thin-commission model follows from the same logic: when software does the work a commissioned human used to do, the margin that funded that human belongs back in the borrower’s price.
Prices are made, not found — assembled from curves, spreads, and incentives that reward whoever understands the machinery.
That is what rates desks taught me. Mortgage borrowers have been on the wrong side of that asymmetry for generations. Closing it is a solvable engineering problem, and it is the one I intend to spend this decade on.