What is a factor model, in plain English

What is a factor model? Strip away the priesthood and the Greek letters and it is a recipe for ranking a pile of investments by qualities you actually care about. That is it. The finance industry has spent decades wrapping this simple idea in robes and incense so it sounds like something only a hedge fund with a server farm could attempt. Nonsense. You already do a scrappy version of it in your head every time you say “this one looks cheap and that one looks shaky.” A factor model just writes that thought down and applies it to a hundred things at once without getting bored or playing favourites.

So let us build the idea up from nothing, one honest piece at a time, until “factor investing” stops sounding like a spell and starts sounding like common sense.

A factor is just one opinion, turned into a number

Picture you are buying a second-hand car. You do not stare at the whole car and emit a vibe. You break it down. How many miles on it? How old? Any rust? Each of those is a single, separate question with a measurable answer. A factor is exactly that: one question about an investment, turned into a number you can line up against every other investment.

In a portfolio the questions sound like “is this cheap relative to its earnings?” or “has its price been climbing steadily?” or “is the underlying business actually profitable?” Each one is a factor. A valuation factor, a momentum factor, and so on down the list. One idea becomes one signal becomes one column of numbers.

The catch is that these numbers arrive in wildly different units. Mileage is in thousands. Age is in years. Rust is a yes or no. You cannot add forty thousand miles to seven years to one rust and announce a total, because the result is gibberish. Same problem with investments: a price-to-earnings ratio and a twelve-month return and a profit margin do not share a language. So before a factor is allowed to play, it has to be made comparable.

Normalisation is the boring step that makes the magic legal

Every flashy tutorial skips this bit, and the whole building leans on it. Normalisation just means putting every factor on the same scale so they can be compared without lying.

Back to the cars. Instead of raw mileage, you ask “how does this car’s mileage rank against the others on the forecourt? Top ten per cent? Bottom half?” Now mileage and age and everything else can be expressed the same way, as a rank or a score, and suddenly adding them up is a sensible thing to do rather than a category error. That translation step is normalisation, and without it a factor model is just a heap of incompatible numbers wearing a confident expression.

Strata makes every factor declare how it gets normalised before it is allowed near a sum. Maybe a percentile rank, maybe a z-score with a clamp, maybe a simple threshold. Once everything speaks one dimensionless language, the maths is finally honest. We went deep on why this matters in factors without black boxes, so I will not relitigate it here. Just know that the dull step is where the real work happens.

A model is the recipe that decides what matters most

One factor on its own is a single opinion, and single opinions are easy to fool. Cheap can mean a bargain or a business quietly dying. Rising price can mean momentum or a bubble inflating. So you stop hunting for the One True Factor and combine several instead, until their blind spots cancel out.

A model is how you combine them. You pick your factors, then you decide how much each one counts. That weighting is the entire personality of the model. If you care most about not overpaying, you lean your weights towards valuation. If you would rather ride what is already working, you tilt towards momentum. Same ingredients, different recipe, completely different ranking falls out the other end.

Here is where most tools quietly betray you. They hand you a single composite score and never tell you who chose the weights or why. Maybe somebody picked them on purpose; maybe they just accumulated like junk in a drawer. Either way you are not allowed to look, so you nod along at a number you cannot question. A model you cannot inspect is just a stranger’s hunch in a nicer font.

Strata ships a trend and momentum battery as a starting point, and you might assume it is The Official Answer carved into stone. It is not. It is a seed. Clone it, change the weights, drop a factor you do not believe in, add one you do. The recipe is yours to argue with, which is the only kind of recipe worth trusting.

A grid is the model let loose on a whole universe

So far we have a factor (one signal) and a model (a weighted blend of signals). The last piece is where it gets genuinely useful. A grid is your model run across a whole universe of instruments at once, then laid out as a view you can actually read.

Think of the forecourt again, except now it is every car in the country, scored by your recipe and sorted best to worst in a single glance. That is a grid. A universe of instruments goes in, the model scores every one of them, and out comes a wall of cells coloured from weak to strong. Your eye finds the strong band before your brain has finished its coffee. This is the part of factor investing that feels like a superpower, because you are evaluating a hundred things with the same disciplined yardstick instead of falling for whichever ticker shouted loudest this week.

The grid never makes the decision for you. It does the tedious, even-handed scoring so you can spend your judgement where judgement actually belongs.

Why you should always be able to open the box

Here is the line that separates a tool you can run money with from a slot machine with good branding. When a cell surprises you, and one eventually will, you should be able to ask it why and get a straight answer.

In Strata every cell in the grid decomposes. Click it and it opens up: each factor’s raw value, what it normalised to, the weight it carried, and exactly how much it pushed the final score up or down. Nothing stays sealed, and you are never asked to take a colour on faith and quietly hope. A score you can take apart is a score you can believe. The other kind is just somebody else’s confidence, borrowed at interest.

That is the same instinct that runs through the rest of the workbench. It is the loop of writing down your reasoning and then checking it against what really happened, applied to ranking instead of to theses. Reasoning you can audit beats a verdict you have to swallow.

So that is the whole of it, with the robes off. A factor is one opinion turned into a comparable number, normalisation makes a pile of those numbers safe to add up, and a model is just the recipe that decides how much each one weighs. Run that recipe across a universe and you get a grid, the whole field scored at a glance. The only version worth using is the one that lets you open every cell and check the working. If you are assembling your own kit for this, our buyer’s guide and our DIY investor tool stack cover where a transparent factor engine fits. When you are ready to stop trusting black boxes and start asking your numbers questions, build your portfolio and open the box yourself.