Where we stand

Schools of Quant

Quantitative research is not one discipline. It is several, and they disagree about something fundamental: what you are allowed to assume comes apart.

This page maps those schools honestly, then says plainly which one BunnyQuant is not - and why that was forced on us by who reads it, not chosen for taste.

The map

Two things place a school: how many instruments it holds at once, and how long it holds them. Almost every professional approach lives on the right-hand side of this chart, because that is where the mathematics works.

Instruments held at onceHolding horizon11000+MicrostructureSignal miningTrendCross-sectionalRisk premiaBunnyQuantone instrument, one decision

The right-hand side is not a fashion. Holding many positions is what makes single-name mistakes cancel out. That is the engine underneath most institutional alpha, and it is unavailable to a person trading one chart.

Cross-sectional factor
Hunts · Ranking thousands of names at one moment
Assumes · Returns separate into independent factor exposures
Needs · Thousands of positions, so single-name errors cancel
Trend / managed futures
Hunts · Persistence along the time axis
Assumes · Momentum is stable long enough to be estimated
Needs · Dozens of markets, long horizons, patience through drawdown
Microstructure
Hunts · Intent in the order flow, not value
Assumes · Queue position and latency are the edge
Needs · Co-location, and a fight you cannot win from a laptop
Signal mining
Hunts · Patterns found by search, not by hypothesis
Assumes · Validation can substitute for theory
Needs · Enormous data, and ruthless discipline against self-deception
Risk premia
Hunts · Payment for bearing a risk others avoid
Assumes · The compensation is structural, not a mispricing
Needs · Years, and the balance sheet to sit through them

The real dividing line

It is tempting to sort quant into "statistical" and "technical", or "academic" and "practical". Neither cut explains anything. The cut that does is this: what does the method assume can be separated?

Factor models assume returns split into independent parts that add back up. Trend models assume the time axis separates from everything else. Both are reductionist in the exact sense the natural sciences use the word, and both are powerful - as long as the parts really are independent, and as long as you hold enough of them that your errors average away.

The other tradition says the interaction is the phenomenon. Fat tails, long memory, volatility that clusters, timeframes that pull on each other, states that shift. Pull those apart and the thing you wanted to measure is gone.

The first tradition has thousands of positions to hide its errors in. The second has one path, and nowhere to hide.

Why a person cannot borrow the institutional answer

A fund holding three thousand names does not need to be right about any one of them. It needs to be right on average, and diversification converts a thin edge into a survivable one.

You hold one instrument. Sometimes two. There is no average to be right about, and no portfolio to absorb a bad call. The mathematics that makes cross-sectional alpha work is simply not available at that size - not because you lack the software, but because you lack the N.

So the question has to change. Not "which of these three thousand names is cheapest", but "what state is this one instrument in right now, and which plan survives that state".

When the models break, they break together

Every few years a move arrives that the prevailing models called impossible. The pattern is old enough to have a shape.

1998
LTCM collapse
assumed: correlations are stable
2007
The quant quake
assumed: our position is ours alone
2008
Credit crisis
assumed: these loans are independent
2010
Flash crash
assumed: liquidity is always there
2018
Volatility shock
assumed: low volatility persists
2020
Pandemic shock
assumed: diversification is enough
2021
Coordinated retail flow
assumed: participants act independently
2022+
War & energy shocks
assumed: geopolitics is noise

What failed in each case was rarely the arithmetic. It was an assumption underneath it: that the pieces were independent, and that the relationship between them measured over calm years would hold through a violent week. When the pressure came, correlations that had sat near zero for a decade went to one - and diversification, the entire engine of the first tradition, stopped working at the exact moment it was needed.

Information cascades

News does something distributions do not expect: it makes everyone update on the same signal at the same instant. A headline out of a war, a central bank sentence, a default - and every participant revises in the same direction together. Positions that were independent by construction become one position by behaviour. This is why the worst days are not merely large; they are large everywhere at once.

The black swan is the most borrowed idea in this history, and the most misused. It was meant as a warning about the limits of induction. It is now mostly used as an excuse - the move was unforeseeable, so no one is answerable for it. Both readings share a flaw: neither is measurable, so neither changes what you do on the day.

"Six sigma" is a statement about your model, not about the market

A move is only extreme relative to the distribution you assumed. Under a textbook bell curve, a 1998-scale move was a once-in-many-lifetimes event; under the market's own measured history, moves of that size arrive often enough to plan for. The number does not describe the world. It describes the gap between the world and the model.

the "impossible" movesize of move →assumed distributionmeasured behaviour

Assumed distribution against measured behaviour - the gap is the whole problem

What BunnyQuant does differently

Tails are measured, not inferred
Risk comes from the instrument's own history of extremes, not from a variance that assumes a symmetric bell. A move that a bell curve calls impossible is simply inside the range we already measured.
Timeframes are not assumed independent
During a cascade they synchronise - every horizon moves the same way at once. We measure that agreement rather than assuming it away, which also means we can tell you when unanimity is conviction and when it is panic.
The system refuses rather than guesses
When price has travelled far beyond its own normal range, the levels below it describe a market that no longer exists. Rather than print them anyway, the product says so, and withdraws any plan whose exit the market has already passed.
Scale is stated in the instrument's own units
Not "six sigma" - which needs a model to mean anything - but how many times its own normal daily move this is. That number is comparable across instruments and does not depend on a curve being true.
What this is not
None of this predicts a cascade. Nobody does. It will not fill you at a level the market gapped straight through, and it will not keep a position safe through a shock. What it does is refuse to present a plan built on a market that has already gone - which is the failure mode that turns a bad day into a ruinous one.

Momentum, and why a reading is not a state

Momentum is the most studied and most crowded edge in the business. It is also where the difference between the two traditions is easiest to see.

Read as a number, momentum says the same thing in both charts below: price has been rising for some time. Read as a state, the two are opposites - one market is extending a move, the other is running out of buyers at the top of one. The number cannot tell them apart. The state can.

Extending
Identical momentum reading
Exhausting
Identical momentum reading

This is why BunnyQuant never presents an indicator reading as a verdict. A reading is an input to a state, and it is the state that decides what a plan is worth.

The three questions we actually ask

Not a scanner over thousands of tickers. One instrument, three questions, in this order - because each answer is meaningless without the one before it.

1 · What state is this timeframe in?

For any interval - 5m, 15m, an hour, a day - the market is somewhere in a repeating structure: building a base, trending up, topping out, or breaking down. We do not label it. We give the probability of each, because on most bars the honest answer is "mostly this, partly that".

12%
58%
21%
Accumulation · 12%Mark-Up · 58%Distribution · 21%Mark-Down · 9%

One timeframe · four states · probabilities that must sum to certainty

2 · Do the timeframes agree?

A five-minute chart in Mark-Up inside a daily chart in Distribution is a different trade from the same five minutes inside a daily Mark-Up. The same setup, the same entry, two entirely different expectations - and the difference is invisible if you only look at one chart.

So we run every timeframe, and then measure how strongly they agree. Agreement is the number that turns a setup into a position size.

1m
5m
15m
1h
4h
1D
1W
AccumulationMark-UpDistributionMark-Down
The short frames lean Mark-Up; the daily and weekly lean Distribution. Agreement is low - and that is the information, not the absence of it.

Seven timeframes · the same four states · one agreement score

3 · Market order, or limit order?

This is the question retail platforms never answer, and it is where money is actually made and lost. A market order fills now, at a worse price, with certainty. A limit order fills at your price, or not at all.

Given the state and the agreement, those two choices have different odds and different risk. We compute both, so the decision stops being a habit and becomes a comparison.

Market order
FillsCertain
EntryWorse
Risk per unitWider
Reaches target first

You pay in price for certainty. Reasonable when agreement is high and missing the move is the bigger risk.

Limit order
FillsUncertain
EntryBetter
Risk per unitTighter
Reaches target first

You pay in certainty for price. Reasonable when the state is ambiguous and patience is cheap.

Same instrument · two ways in · two different sets of odds

The same tools. A different question.

Nothing on this page is an argument against rigour. We use the machinery the industry uses - and we hold ourselves to the same tests.

Data science
Every claim is measured, not asserted. Anything that cannot be checked does not ship.
Scale
Years of history across equities, futures, crypto and Vietnamese markets, on seven timeframes, refreshed continuously.
Out-of-sample
Fitted on one period, judged on another it never saw. A result that survives only in-sample is not a result.
Sharpe and its limits
We report risk-adjusted return, and we say where it misleads: Sharpe treats a bad month and a ruinous week as the same kind of event. Tails are measured separately.
Stated uncertainty
Every output carries a confidence. When the evidence is thin, the product says so rather than rounding it into a signal.
What we do not claim
No forecast of price. No guaranteed win rate. No suggestion that a probability removes risk.

In one line

The other schools ask what explains the average return of many instruments. We ask what state this one instrument is in, and which plan survives that state. Those are different questions - not competing answers.

How it is builtSee plans

Analysis, not investment advice. Nothing here is a recommendation to buy or sell, and no probability removes the risk of loss.