QuantConnect.
The host firm lived for 13 years selling backtest infrastructure. Its one bet on selling signals, Alpha Streams, closed in four. The sharpest evidence that the signal isn't the product.
A platform that knew exactly which layer it was at.
QuantConnect was founded 2011 by Jared Broad. The product was, and is, a hosted quant infrastructure stack: a backtesting engine (LEAN, open-sourced 2014), unified data feeds, broker integrations, live execution, paper trading, a research environment. Everything except the signal itself.
The host firm built a defensible business inside the infra layer. LEAN became the most-used open-source backtester in retail and small-shop quant. Subscriptions to the hosted version compounded. The firm grew without ever needing to take a view on what to trade.
Then, in May 2018, QC launched Alpha Streams, a marketplace where the platform's algorithm-writers could sell their signals to institutions. Same product idea every alpha marketplace has had since Quantopian and Numerai. Discover the next great strategy. Different pitch, same architecture: signals as the unit of value, marketplace as the fusion layer. Four years later, it was gone.
Infra lived. Signals died. Same company.
QuantConnect is the cleanest controlled experiment we have for take C. One firm ran two products at once. Only one survived.
QuantConnect founded
Hosted quant research platform: backtesting, data, paper trading. Initial product targets indie quants and small shops priced out of Bloomberg / FactSet stacks.
LEAN open-sourced
The backtesting engine goes on GitHub. Over five years it becomes the default open-source backtester for retail and academic quants. Stars, contributors, integrations into broker APIs.
infra · network effects compoundLive brokers · IBKR, OANDA, GDAX
Platform extends from backtest into live execution through partner brokers. Still selling infra, still no view on what to trade.
Alpha Streams launches
"Discover, license, and trade quantitative strategies built by our community." The marketplace pitch: algo-writers publish signals, institutions license them, QC takes a cut. Same model as Quantopian (then alive), Numerai, and dozens of forgotten precursors.
pivot · signals as the unit of value~150 active alpha contributors · long-tail revenue
The marketplace fills with strategies. Most receive zero subscriptions. A handful do well. Even for those, every institutional buyer is essentially trying to build their own internal combinator on top, which means they're not paying for the signal, they're paying for the data.
Quantopian closes the equivalent program
Quantopian, the better-funded competitor, shuts its crowdsourced fund Oct 2020. The signal-marketplace category loses its most-cited example. QC continues with Alpha Streams.
Alpha Streams shut down
QuantConnect announces wind-down. Stated reasoning: "We will be focusing on our core product." Translation: the infra business is the business; the signal marketplace was never the moat.
4 years · marketplace closed · host firm continuesHost firm · year 13 · still selling infra
QC continues as the standard hosted quant platform. The two-product experiment is over. The signal layer was a four-year detour; the infra layer is a 13-year compounding business.
infra: 13 yrs · signals: 4 yrs · controlled experiment endedEvery institutional buyer who walked through Alpha Streams was, in effect, building their own combinator on top. They weren't paying for the signal, they were paying for the chance to fuse it. The marketplace had picked the wrong layer to sell.From the StratCraft signal-fusion thesis · "the signal isn't the product"
A marketplace can't substitute for fusion.
Read at the combinator layer: Alpha Streams sold the matching market between signal producers and signal consumers. But a market doesn't fuse. The buyer still had to bring the combinator. Eventually the buyer realized that, and stopped paying.
Marketplace as fusion · doesn't fuse.
Algo-writers ship signals, institutions browse and license. The marketplace runs the matching layer. No one runs the fusion layer: the buyer has to bring it themselves. So the buyer pays for data, not for fusion.
The product is the combinator itself.
The customer doesn't license signals to fuse later. They install a combinator that does the fusion. Five methods, confidence-weighted, ranked scoreboard. The thing the marketplace asked the buyer to build, productized.
Three things Alpha Streams proved.
The signal isn't the unit of value.
Four years of Alpha Streams couldn't convince institutional buyers to pay for signals as a finished product. They kept paying for the data, never the decision. That's the price ceiling on a signal sold à la carte.
↪ Read the taxonomyA market isn't a combinator.
Marketplaces match supply and demand. They don't weight, fuse, or schedule. Every buyer at a signal marketplace ends up rebuilding the combinator anyway, at which point the marketplace becomes a commodity feed.
↪ See the architectureSell the layer that does the work.
The same firm that succeeded selling infrastructure failed selling signals. The difference: infrastructure does work the buyer can't replicate. A signal is just a number. The buyer needs the combinator to make it actionable, and they'd rather own that than rent it.
↪ View the infographicFour firms. One pattern.
Sell the combinator, not the signal.
Markets don't fuse. Combinators do.