SimicX
ALPHASTREAM · PRODUCT BRIEF

Read the world’s research.
All of it.

Alpha often hides in research nobody else has read. New studies appear every day, scattered across languages, markets and disciplines. No research team, however large, can read them all, let alone test them. SimicX’s AI scientists read the world’s research, break it into testable ideas, fuse those ideas into new models and let the real market grade them. Your team starts from what has already been proven.

Reads everythingBreaks it downFuses the partsTests hard
SIMICX FIRST-PARTY DATAILLUSTRATIVE
Of every 100 studies, about 3 show real-world impact
6Mpieces of research distilled so far
3%show real-world impact
30xresearch efficiency for clients
Why 30x: start from the 3% that work and you skip the 97% that do not.
01 · THE PROBLEM

Traditional research hits three walls: you can’t trust it, you can’t keep it, and you can’t afford to test it

Most research does not survive scrutiny. Good ideas start to decay the moment they go public. And testing them one by one, by hand, costs too much.

82%

Can’t trust it

Under stricter multiple-testing standards, 82% of 452 market anomalies fail.

FailHold

Copy a study straight into a portfolio and you are probably trading noise.

Hou, Xue & Zhang · Review of Financial Studies · 2020 ↗
58%

Can’t keep it

Once a factor becomes public, its portfolio returns fall by 58% on average. The window starts closing the day an idea goes public.

Before
After

Whoever reads and tests an idea first gets the most out of it.

McLean & Pontiff · Journal of Finance · 2016 ↗
40hrs

Can’t afford to test it

On average, it took 40 hours to complete the initial research on one good alpha signal.

That means a quant can develop only a few alphas a month.

SimicX · our own empirical testing
There are always more ideas than people to read them, take them apart and test them. AI can do that work.
02 · WHY A GENERAL MODEL IS NOT ENOUGH

Your team already has an LLM. It still needs a research machine.

A general model can summarise a paper and write its code. What it cannot tell you is whether the code is right, whether the backtest is clean, or what was already tested last month. SimicX runs thousands of AI scientists through one research loop around the clock, with those checks built in.

<40%

The code can be wrong

Even the best of more than 30 AI models tested implemented less than 40% of the code from recent research papers correctly.

Code that runs is not the same as code that is right.

Hua et al. · ResearchCodeBench · arXiv 2506.02314 · 2025 ↗
51–62%

The backtest can be contaminated

Backtested Sharpe ratios of LLM trading agents fall by 51–62% once the test moves past the model’s training cut-off. The model remembers the market.

A backtest inside the model’s memory proves nothing.

Li et al. · Profit Mirage · arXiv 2510.07920 · 2025 ↗
A general model answers one question at a time. A research machine reads everything, tests every idea the same way and keeps every result.
01 · READS EVERYTHING

Full coverage, no blind spots

New working papers and preprints are read within days of release. Journals, classic texts, conference notes, broker research, market analysis, news and fields outside finance are all in scope. Everything is read in its original language.

02 · BREAKS IT DOWN

Split into testable ideas

We take each piece of research apart into separate ideas and keep only the sound ones. Every idea links back to the exact passage it came from.

03 · FUSES THE PARTS

Fusion across studies and sciences

Strong alphas usually come from combining ideas. The AI finds ideas that fit together across studies that never cite each other and fuses them into a new trading rule. The space of combinations is far too large for people to search.

04 · TESTS HARD

Strict tests in a sealed sandbox

A second AI plays the sceptic and pokes holes. The code runs in a sealed sandbox: each day it sees only that day’s real market, so nobody gets to peek at tomorrow; each year it learns only from the years before; every trade pays its costs. After the backtest come a sealed test and a live forward test.

05 · KEEPS EVERY RESULT

Every test becomes knowledge

Pass or fail, each result flows back to every idea involved: what works is reused, what fails is recorded, and the same test never runs twice. The system gets smarter the longer it runs.

CROSSING FIELDS

Breakthroughs often come from another field

The Black-Scholes option pricing formula grew out of the heat equation in physics. A study of 17.9 million papers found that research adding unusual combinations to conventional work is

nearly 2x

as likely to become a high-impact hit. A person masters one or two fields at best; AI can span them all.

Uzzi, Mukherjee, Stringer & Jones · Science · 2013 ↗
The research universe is full of noise You do not want to spend your team’s hours, tokens and compute on noise. Let us do that for you. We sift all of it, and you start from the research that held up.
03 · A REAL EXAMPLE

How an alpha is made: six studies from all corners of science

This alpha runs on AlphaStream today. The AI took the best parts of six unrelated studies and fused them into one model. It is one configuration of the machine: S&P 500 stocks, rebalanced weekly.

THE STUDIES6
WORKING PAPER
Interpretable Systematic Risk around the Clock
PROCEEDINGS
Adaptive Long-Short Equity Strategies with Salience Theory and Hidden Markov Regimes
PREPRINT
Demystifying the Trend of the Healthcare Index
JOURNAL · THAI MARKET
Identifying Risk Regimes in a Sectoral Stock Index
JOURNAL · SWEDISH MARKET
MAX Effect and Investor Sentiment: Evidence from the Swedish Stock Market
OUTSIDE FINANCE · AGRICULTURAL ECONOMICS
The Impact of Phosphate Fertilizer Industry Consolidation on Future Phosphorus Supply
THE IDEAS WE TOOK5 of 13
1
Market risk by time of dayOvernight and daytime risk earn different rewards
2
The shape of a day from four pricesOpen, high, low and close split a day into overnight and daytime moves; found in two studies
3
The “lottery stock” measureThe average of a stock’s best few days: a steadier version of a known warning sign
4
Investors notice what stands outWeight each day by how far it stood out from the market
5
A market-concentration indexBuilt to measure how concentrated fertiliser supply is; here it measures how bunched a stock’s gains are
8 other ideas from these studies were not needed here, such as a news-reading model and a commodity trade model.
THE ALPHA1
Overnight-Risk Lottery-Guard
Favours stocks that move with the market overnight more than by day, and avoids stocks whose recent gains look like lottery tickets.
LEG 1 · IDEAS 1, 2When does the stock take its risk? Stocks that move with the market overnight more than by day score higher.
LEG 2 · IDEAS 2 TO 5Does it look like a lottery ticket? Big, bunched, eye-catching up-days score lower.
Each week: rank the S&P 500, go long the top 10% and short the bottom 10%.

The value of fusion: information ratio up from 1.21 to 1.55

Idea 1 alone
1.21
Ideas 1 to 5 fused
1.55

Information ratio: return per unit of risk. Backtest: point-in-time S&P 500 constituents, weekly rebalancing, about five and a half years; walk-forward by year, learning only from the past, after trading costs. Past performance is not a guide to future returns.

Failure is knowledge too Not every fusion works. In another case, the new ideas added nothing, so the system recorded that and tried them in other combinations. Applying a fertiliser-industry measure to stocks is a leap few researchers would ever try; AI can try leaps like it systematically and at scale.
04 · HOW WE WORK WITH YOU

The machine runs to your specification, and you keep everything

We work with you to define the specification: the trading universe, the trading parameters and the backtesting method, down to the bar an alpha has to clear. The machine then searches inside those limits.

RUN SPECIFICATIONFOR ILLUSTRATION ONLY
UNIVERSE
US stocks, index members as they stood on each date
REBALANCING
Weekly
PORTFOLIO
Long-short, top and bottom 10%
RESEARCH LINES
Momentum, liquidity, profitability
DATA
Prices, fundamentals, filings and your own data
BACKTEST
Walk-forward by year, after trading costs
BAR TO CLEAR
Information ratio above 1 after costs, out of sample

Every line is yours to change: a global universe, monthly or intraday rebalancing, a long-only or multi-factor portfolio, your own data on your own servers. Coverage follows the data: stocks and crypto today.

What comes back
  • The source code of every alpha that clears the bar
  • Its backtest and full record: thesis, sources, construction and risks
  • A live rank list on the schedule you set
  • All of it by dashboard, API or SFTP, or simply as Excel files

You run the machine on your terms, and everything it finds is yours.

05 · FOR DISCRETIONARY TEAMS

AI does the reading and testing; the PM makes the call.

Quant or discretionary, you get a research team that never tires, reads research from the whole world and shows the source for everything it finds.

EXPLAINABLENo black box: every claim has a source

Each alpha comes with a full record: thesis, mechanism, economic rationale, “who pays for this return”, construction parameters and risk warnings. Every idea traces back to its source.

USEFULSee the corners your team can’t reach

The newest and most obscure research worldwide becomes input you can debate and act on, across markets, languages and fields your team has no time to cover.

TESTABLELet the market score it before you size up

Does a hunch actually work? Let IC, ICIR and quantile returns answer. Before real money goes in, historical and live data deliver the verdict.

Leave to AI the research no team has time to finish. Leave to the PM the calls only a PM can make.The SimicX product philosophy
06 · WHAT THE MACHINE HAS BUILT

AlphaStream: every alpha the machine builds, open to inspection

SimicX is an alpha generation machine, not an alpha vendor. AlphaStream is where you inspect its work: every alpha searchable, comparable and traceable, with live signals and metrics by dashboard and API. The full system can also be deployed on your servers, where it works with your data, your alternative data, your signals and your research.

simicx.com/platform
The AlphaStream console: a searchable table of alphas, with one row open to show its three key charts 1 2 3 4
1
Filter like a search engineType a query such as icir>1 freq:weekly and the signals you want appear at once.
2
Key metrics on one screenMean IC with Newey-West t-stat, ICIR and IC hit rate, long-short and long-only return and max drawdown, turnover.
3
The three charts that matter mostOpen any row to see the weekly ICIR moving average, decile excess returns and cumulative IC.
4
Full record, side by sideOpen an alpha’s full record in one click; tick two or more to compare them.
INSIDE ONE ALPHA

Open any alpha and it stands up to questions

Shown here: Overnight-Risk Lottery-Guard, the alpha from the example above. Click any image to enlarge.

AlphaStream · full record
An alpha's full record: thesis, construction and research sources
Full recordThesis, construction and sources on one page.
AlphaStream · predictive power
Predictive power: weekly ICIR, decile excess returns, cumulative IC and the compounded top-minus-bottom spread
Predictive powerICIR over time, decile excess returns, cumulative IC and the compounded top-minus-bottom spread.
AlphaStream · orthogonality
Orthogonality: return, signal and stress correlation and the effective number of bets across three alphas
OrthogonalityReturn, signal and stress correlation, plus the effective number of bets: see at once whether a new signal is really new.
RESEARCH TRAIL

An idea can come from any market, in any language

The research behind an alpha does not need to come from its home market. Two of the four sources behind a later version of this alpha are broker reports on another market, which the AI read in their original language. The ideas distilled from them ended up in a US equity portfolio.

Which study, which idea, which leg: every step is on the record.

An idea can travel across markets
while the data stays where it is.
AlphaStream · research trail
Research trail: two broker reports in another language and two English studies, five ideas taken from them, and the three-leg alpha they became
Point-in-time data, no look-aheadEach trading day uses only the index members and prices as they really were, which rules out look-ahead and survivorship bias.
Walk-forward, out of sampleEach year learns only from the years before, then proves itself in years it has never seen.
Returns after trading costsEvery trade pays its costs and turnover is in plain sight, so paper returns stay close to reality.
Portfolio-level comparisonSignal, return and stress correlation plus the effective number of bets show whether a new signal is really new.
Full record, exported to PDFThesis, mechanism and risk warnings, down to “what these numbers do not say”, all spelled out.
Always running, always currentAlphas that pass move to AlphaRunner and keep running on new data; signals and metrics arrive by dashboard and API.
WE RUN IT FOR YOU

AlphaStream

We run the machine to the specification you set. The alphas that clear your bar reach you by dashboard and API, each with its full record. Licensed per team and per asset class.

ON YOUR SERVERS

AlphaCoder

The full system, deployed on your own servers. It works with your data, your alternative data, your signals and your research, so every alpha it generates is unique and belongs to you. It builds a book for you and shows it only to you.

Pricing: a licence fee plus a share of the profit, so our interests are aligned with yours.

ABOUT SIMICX

So every R&D team builds only on research that works

SimicX is a London-based AI lab building the verification layer for science: AI scientists that test whether research holds up in the real world.

Dr. Simon WangCo-Founder · CEOPhD in applied mathematics; founding CTO at Stable, taking it from 0 to 1 and from London to global. Builds models from research and puts real money behind them.
simon@simicx.comLinkedIn ↗
Rachel QianCo-Founder · Quant LeadTop quant, Oxford MSc with Distinction. Eight years of hard maths with Simon at Stable: $60M+ raised, a product launched on the CME and sold worldwide.
rachel@simicx.comLinkedIn ↗
Dr. Jamin WangCo-Founder · Commercial LeadPhD in electronic and electrical engineering; ACI, then Alipay, then scaled Alibaba Cloud as Head of Northern Europe, doubling ARR every year for three years straight.
jamin@simicx.comLinkedIn ↗
40+financial institutions in conversation with us
For the internal use of institutional investors only. Not investment advice, an offer or a recommendation. Backtest results are based on historical data; past performance is not a guide to future returns.
Figures are SimicX first-party data or cited research. © SimicX · London · simicx.com
SimicX