He ranked the people whose job is to rank everyone else. Then he built a company out of the answer.
The look of a man checking your backtest. Vinesh Jha, who would like to see your point-in-time data before he believes the chart.
Most people in finance fall for a good story. Vinesh Jha spends his days proving the story wrong. His firm, ExtractAlpha, takes the raw and the unglamorous - earnings transcripts, web traffic, the chatter of message boards, the timing of an analyst revision - and asks one stubborn question of each: does this actually predict anything, or does it just sound like it should? The answer, far more often than the data vendors selling it would like, is no.
That skepticism is the product. ExtractAlpha launched in 2013 out of Hong Kong with a deceptively simple pitch: data could be packaged into clean, actionable signals for institutional investors, and someone had to do the boring, rigorous work of separating the predictive from the merely interesting. More than a decade later the firm is still independent, which in the boom-and-bust world of alternative data is roughly as common as a hedge fund admitting it was wrong.
Today Jha runs a team scattered across Hong Kong, the United States, Europe, and Canada, building signals that hedge funds and institutions trade on. He also runs Estimize, the crowdsourced earnings-estimates platform that merged with ExtractAlpha in 2021 - a marriage of the wisdom of the crowd with the discipline of the quant. The two halves fit his worldview perfectly: gather a lot of opinions, then test which ones are worth anything.
He did not arrive here by accident, and he did not arrive in a hurry.
Signals require rigorous testing, point-in-time data, and economic intuition rather than compelling narratives.
— Vinesh Jha, on what actually counts as a signalBefore the markets there was the math. Jha did his undergraduate degree in mathematics at the University of Chicago and a graduate degree at the University of Cambridge - pure mathematics, the kind with no obvious cash value attached. The translation to finance came at StarMine in San Francisco, where from 1999 to 2005 he was Director of Quantitative Research and built the factor models that let big investors do something surprisingly hard: tell which Wall Street analysts were actually good.
It is a delicious problem. The whole industry is built on analysts issuing forecasts, and almost nobody had a rigorous way to score them. Jha built one. His metrics of sell-side analyst performance became industry standards, and the alpha signals he built on top of analyst, fundamental, and other data turned into real commercial products.
Then he crossed to the other side of the trade. He developed systematic strategies on the proprietary desks at Merrill Lynch and Morgan Stanley in New York, and eventually landed as an Executive Director at PDT Partners, the famously selective quant shop spun out of Morgan Stanley's premier prop-trading group. This is the part of the resume that makes other quants sit up: PDT is where careers usually peak, not pivot.
Jha pivoted. The trigger was partly mundane - a non-compete clause after he left PDT, the kind of contractual pause that has launched more than one company. But there was a deeper itch. He has talked about how once an organization grows past 100 to 150 people, it stops feeling like a startup. He wanted the startup feeling back. So in 2013 he built one.
When a dataset is fashionable, it is usually already priced in. Jha argues the unfashionable corners - earnings forecasting, company fundamentals - keep working precisely because everyone has wandered off to chase the new shiny thing.
A signal earns its place through point-in-time data and economic intuition, not a compelling backstory. If it only works when you peek at data you would not have had at the time, it does not work.
He chose to bootstrap and keep control rather than raise heavily. Slower, harder, more personal effort - and entirely his. A decade of independence is the proof of concept.
Pick a project you're passionate about and try it - you're going to screw it up, and it's better to do that on your own project than someone else's.
— Vinesh Jha, advice to young quantsExtractAlpha's edge is partly geographic and partly linguistic. Headquartered in Hong Kong with native speakers on staff, the firm runs natural language processing across English, Japanese, and Chinese - mining earnings calls, filings, and sentiment for the signals that an English-only shop in Connecticut would never see. Asia-specific datasets are a deliberate specialty, not an afterthought.
The product line runs to roughly sixteen datasets, including TrueBeats, which forecasts EPS and revenue surprises - a direct descendant of the analyst work Jha started back at StarMine. The throughline of an entire career is right there in a single product: the question of who is about to be wrong about earnings, answered with math.
Then there is Estimize, the other half of his desk. It crowdsources earnings estimates from tens of thousands of contributors - independent analysts, students, professionals - on the wager that a diverse crowd, properly weighted and de-biased, beats the narrow Wall Street consensus. Pairing it with ExtractAlpha's quant machinery is the whole thesis made literal: collect many opinions, then ruthlessly test which ones to trust.
In early 2026 he was still publishing on the theme that haunts every data firm - how to stay differentiated when everyone has access to the same feeds. His answer has not changed in thirteen years. Do the rigorous, unglamorous work that nobody else wants to do, and do not mistake a good chart for a good idea.
He was a pure mathematician before he ever touched a trading desk. The markets were the application; the math came first.
His StarMine models scored the Wall Street analysts whose forecasts the rest of the industry was busy quoting.
Hong Kong, the US, Europe, and Canada. The sun rarely sets on a running ExtractAlpha backtest.
The pause after leaving PDT Partners gave him the runway to start building ExtractAlpha.