InvestYog is a research and education desk for learning how AI and quantitative methods are applied to markets — and, more importantly, how they are tested. Every call it makes is written down before the outcome is known, scored against real prices afterwards, and kept whether it was right or wrong.
Research and education only — not investment advice, and not a place to trade.
The same four steps run every trading day. The order matters more than any individual step: the call is fixed before anyone knows the answer, which is the only way the record means anything later.
Ten minutes before the US market opens, three independent models each rank the covered universe and take their single best name. Entry, target and stop are sized from that stock's own recent volatility. The picks are then fixed for the day.
Once the session ends, every open pick is marked against the day's actual closing price — ahead of its entry or behind it. The verdict is written once and never revised.
A pick ends when its target or its stop is reached, or when its holding horizon expires. That outcome — not the daily mark — is what counts toward a model's record.
Hit rate, average return and average miss accumulate per model. A model needs a real sample and has to beat the rate you would get with no skill at all before its record counts for anything.
These are the concepts the app is built around. They are also the ones that separate a result you can rely on from one that only looks convincing.
A stock's move is partly the market's move and partly its sector's. Strip both out and what remains — the residual — is what is specific to that company. The claim being tested is that an unusually large residual tends to snap back. Testing it on the raw price instead would mostly measure the market.
A z-score restates a move in units of that stock's own recent variability: −2 means "two standard deviations below its normal", not "down a lot". It is what lets a calm utility and a volatile chip stock be compared on the same scale.
Every published pick carries a target and a stop derived from the stock's own average true range, so a jumpy name gets a wide bracket and a quiet one a tight bracket. A fixed percentage stop would be far too tight on one and meaningless on the other.
A model is trained on the past and tested on the future it has not seen — repeatedly, moving forward through time. A deliberate gap is left between the two so that overlapping windows cannot leak tomorrow's answer into today's training data. Most impressive backtests fail this.
A 62% hit rate sounds like skill until you notice the stock rose on 62% of days anyway. The honest comparison is against the constant-call base rate — what you would score by always saying "up". Measured properly, results that looked significant here stopped being significant.
Try enough parameter combinations and one will look excellent by luck alone. That is why every trial is logged, and why a result is discounted by how many were tried before it. A backtest without that accounting is a story, not evidence.
The full course — price and trend basics, sentiment, portfolio risk, reading filings, and how market makers actually work — is inside the app on the Learning page.
InvestYog is invite-only while it is being built. Every account sees the whole research desk; what a tier changes is how much of the shared compute and data budget an account may spend — model training, live vendor fetches and AI analyses all cost real resources drawn from one pool.
The preview default.
Room to run your own work.
For heavier research use.
The desk's own tier.
Read-first, on purpose. In the current preview the actions that spend the shared budget — adding a symbol, training a model, running an AI analysis — are performed by the desk, so that one member cannot exhaust the day's allowance for everyone. Reading, screening and learning are unrestricted for every account. No pricing has been set; nothing here is an offer.
No. Nothing here is advice, a recommendation, or a solicitation, and none of it is personalised to your circumstances or goals. It is a research and education platform. Every signal, prediction and backtest is a demonstration of a technique. Decisions — and their consequences — are entirely yours, and you should do your own research and consider speaking to someone qualified before acting on anything.
We have no basis to say so, and we would rather tell you that than imply otherwise. The strategy this desk was built to test showed no demonstrated edge once it was measured against the correct benchmark. That result is published inside the app rather than buried. The value on offer is the method — how a claim like that gets tested honestly — not a promise of returns.
No. There is no broker connection, no order routing and no account holding money. It is a research desk, not an execution venue, and it works on end-of-day data that would be unsuitable for trading even if it did.
Prices and fundamentals come from public market data; company financials from the SEC's own filings; disclosures from the House Clerk and Senate systems; headlines from public news sources. During market hours figures are refreshed on a few-minute cadence — deliberately not tick-by-tick, because nothing here is time-sensitive. Every number is stamped with when it was measured.
People who want to understand how quantitative and AI methods are applied to markets, and how such claims are validated — students, engineers, analysts, and anyone tired of tools that show a chart and skip the evidence. It assumes curiosity, not a finance degree; the Learning course starts from first principles.
Treat AI-generated output as a starting point that may be wrong. Where a language model is used here it is grounded in real data — actual filings, actual headlines, actual computed indicators — and it is structurally prevented from stating a price target. Where research gave no basis for a view (say, a one-year direction), the app says "unable to predict" instead of inventing one.
It is an early preview run by a small team, and the data and compute behind it are a shared, finite budget. Keeping the group small keeps the service usable and lets us actually talk to the people using it.
An account is an email address, a display name and a sign-in history. There is no trading account, no brokerage link and no payment information. The company's Privacy Policy, linked below, governs the rest.
Already invited? Sign in and start with today's picks and the record behind them.