How Theta Research analyzes a company — our lens, our rules
Placeholder = ourselves · all six axes at 10 to show what a perfect score looks like · click any axis for the rule behind it
Deep-research methodology
Placeholder: the company's next scheduled catalyst and what it means for the thesis.
Theta's lens · methodology
From raw disclosure to a verdict card
A valuation's verdict rests entirely on its assumptions, and in most research those assumptions are guessed. Theta does not replace the method; instead a deterministic engine supplies its weakest link, the assumptions themselves.
01What valuation is: not the arithmetic, the assumptions
Every valuation, whether DCF or multiples, is at heart a calculator: assumptions about growth, margins, returns on capital, discount rate and persistence go in, and one value comes out. Themath is grade-school; what decides the answer is the handful of input assumptions.
02How institutions commonly value a company
- Intrinsic (absolute): DCF discounts future free cash flows at the cost of capital and adds a terminal value; reverse DCF instead asks what growth the price already implies and whether the business can deliver it; DDM is the dividend special case.
- Relative (multiples / comps): apply peer multiples such as EV/EBITDA, P/E and EV/Sales to the target; it is the day-to-day workhorse.
- Scenario / probability-weighted: give bull, base and bear targets each a probability, weight them, then compare to the price for the margin of safety.
- SOTP, NAV, real options: for conglomerates, asset-heavy or distressed names, and optionality in biotech and commodities.
Strip them down and they share one value trinity
- how fast it grows, and for how long
- whether returns beat the cost of capital (ROIC versus WACC), which decides if growth creates or destroys value
- how long that excess return lasts, the top driver of terminal value
As Damodaran puts it, a valuation is a story about a business, disciplined by numbers.
03Two common approaches, each with real shortcomings
The old buy-side
The old buy-side sets these assumptions by hand: how fast it grows, how long it lasts, what discount rate to use, often drawn from one management meeting, a relationship, or a single person's instinct. It varies from desk to desk, cannot be audited after the fact, and when it is wrong no one can point to which input failed. In the end it is a guess dressed in a polished spreadsheet.
The AI black box
The newer crowd runs to the other extreme: pour every series into one large model and read off a number or a buy-sell call. It cannot say why, it leaves no trail, it quietly overfits its own backtest, and it treats a non-stationary, low signal-to-noise market as if it were image recognition. Worst of all, it removes the one safeguard that should remain: a person who can be held to the call.
Pricing a company and reaching a verdict usually either rests on one person's judgment or is handed to a model no one can explain.
Theta takes a third path: a quantitative model with deep reinforcement learning at its decision core. What code can settle as fact is settled as fact, and the learning layer is used only where the relationship is nonlinear; every one of the six axes enters that model as an input and is analyzed together. Why it is more trustworthy is simple: every input is traceable, every step leaves an auditable trail, and the final call stays with a person who can be held to it.
04How Theta does it
The methods are mature, but the growth, persistence and discount assumptions are mostly set by hand, vary from analyst to analyst, and are hard to audit.
We use a deterministic engine to derive, from raw disclosure (EDGAR, 13F, FINRA), what those assumptions should be, leaving an auditable trail. The six axes are where those assumptions come from.
05The six axes and the valuation knobs
Each axis calibrates one knob of the valuation:
①②③ are the value trinity itself, while ④⑤⑥ are the price-relative layer and the edge. Once the six scores are set, an evidenced set of DCF and scenario assumptions follows: the arithmetic is standardized, and the hard assumptions are already produced.
06The weights are dynamic
How much each of the six lenses contributes to the final read is dynamic. Each lens moves, and so does the weight it carries in the synthesis: change the market regime or the industry, and which lens matters most changes with it. A neural network learns these dynamic loadings, and how the six combine into one score is decided by the supervised and reinforcement-learning layers.
07Quality calibrates the valuation, it does not skip it
① low, ② low (ROIC about equal to WACC), ③ 6, so it is valued on mid-cycle normalized earnings at a low cyclical multiple, with a wide bull-to-bear band and none of a compounder's runway or rich terminal.
a long runway, a rich terminal multiple, a tight band; the same DCF, every assumption lifted by quality.
Same calculator: change the six scores and the value it prints shifts by an order of magnitude.
08The boundary
The six axes will not print a price or set your discount rate; to get a number you still add a standard DCF or multiples shell on top. But that shell is the simplest, most standardized part, and the hard assumptions are already produced deterministically.
Our bet: code computes the facts, plus one auditable judgment, beats any single number or pure narrative.