No daily noise. A new post lands when the experiment changes, the evidence improves or a failure changes what gets tested next.
Update 03
The final exam.
The race went live, and the best ideas faced six months of data they had never seen.
Monday was the first live paper race: Scout traded on its own, and Quant traded in shadow mode. Day trading is still
losing small amounts, and a new test showed it loses about the same in rising and falling markets. The problem is the
approach, not the market.
The research moved to slower ideas that hold stocks for months. The strongest candidates took their one-time final exam
on half a year of locked-away data. The star of the weekend failed outright. One idea, built on what Wall Street analysts
do, held up in a weaker form and now faces the toughest judge of all: the months ahead.
We also tested famous published strategies: classic “cheap stock” and “quality” rules, well-known lazy portfolios, and
copying the quarterly holdings of famous investors. Most didn’t beat simply owning the market. A couple deserve a closer look.
TestedFinal exam on locked data, plus famous strategies
LearnedEven good-looking ideas break on new data
NextForward-test the survivor; check the newest leads
Update 02
Ten thousand ideas. Zero proven.
A weekend of testing everything we could think of, and a few popular bots from the internet too.
Quant’s research lab grew from a few dozen ideas to more than ten thousand tested strategies: price patterns, news, company
finances, insider trades, small companies, time-of-day trades, market timing, pairs and hundreds of crypto coins, across ten years
of history. Every idea was written down before it ran and judged on years it had never seen. Random picks ran alongside as a
fake-strategy check.
None passed. Many looked brilliant on the years they were built on and fell apart on new data, or their edge was smaller than
the cost of trading. We also ran dozens of popular free community trading bots in their own software. None made money on the
newer data, and many lost most of the account even during a huge crypto boom.
A handful of ideas did beat random picks in both periods. They are not proven, so they get one clean, final test before anything
changes how Quant trades.
Tested10,000+ strategies and 40+ public bots
LearnedMost “winners” are luck or costs
NextLive paper race starts Monday, Nov 2
Update 01
The lab is built.
Two bots, years of market history and one hard lesson before the race even starts.
Scout’s rules got their first automated day on pretend money. The two trades lost a combined $14.92, and the safety limits held.
Quant became a separate research system with a historical price database, a record of every signal and shadow-trade experiments.
It has not found a profitable edge yet.
The biggest lesson so far: modeled trading costs erased the starting strategy’s slight advantage. Several other ideas also failed.
The one near-breakeven idea still needs evidence from genuinely new trading days.