How to decides whether a stock is worth buying using AI | paper-profit

paper-profit

How to decides whether a stock is worth buying using AI

It started with a bad idea

When I first started building PaperProfit, one of the problems I had to solve was figuring out how to evaluate a stock.

My first attempt was embarrassingly naive. I just asked an AI: “Is Apple a good stock right now?”

The answer came back confident, fluent, and almost completely useless. The model had no idea what Apple’s current price was, whether the market was up or down, or whether Apple was cheap or expensive relative to its own history. It was like asking someone who’d been in a coma for six months to give you stock tips. The words sounded right. The substance wasn’t there.

So I tried another approach: I found a data feed of Wall Street analyst ratings — the “buy/hold/sell” recommendations from big banks — and averaged them together. Surely the professionals knew what they were doing?

Also useless. Analyst ratings, it turns out, are often outdated, conflicted (banks make money from the companies they cover), and hopelessly optimistic. When every analyst rates everything a “buy,” the signal disappears.

I was stuck. And I started to wonder: how does anyone actually evaluate a stock?

A crash course in how Wall Street thinks

I spent weeks trying to figure this out. I read books, watched finance YouTube, had long conversations with AI assistants, and dug through academic papers on investing.

Here’s what I learned: professional equity analysts — the people at hedge funds and banks whose literal job is figuring out which stocks to own — don’t rely on any single signal. They combine multiple lenses, each answering a different question about a company.

The goal, I realized, shouldn’t be to “predict the market.” It should be to build an educational framework that mirrors how equity analysts actually think.

After simplifying this into something a software project could actually implement, I landed on three core pillars:

The Three-Pillar Framework

Pillar<br>Purpose

Fundamental Analysis<br>Measures financial strength and valuation

Technical Analysis<br>Evaluates price action and momentum

Qualitative Analysis<br>Interprets management quality, risks, and narrative

Pillar 1: The Numbers (Fundamental Analysis)

This is the financial health check. Think of it like looking at a company’s report card.

Is the company making money? Is it growing? Does it have a mountain of debt that could sink it in a recession? Is the stock cheap or overpriced relative to what the business actually earns?

A few of the numbers we look at:

Revenue and growth — Is the company selling more this year than last year?

Profit margins — Of every dollar they earn, how much do they actually keep?

Debt — Could they survive a rough patch, or are they one bad quarter away from trouble?

Valuation — The P/E ratio, in plain English, is how many years of earnings you’re paying for. A P/E of 20 means you’re paying $20 for every $1 the company earns. High isn’t always bad (it might mean the market expects growth), but it does mean you’re paying a premium.

Pillar 2: The Chart (Technical Analysis)

This one is about price movement — not what the company is worth, but what the market thinks it’s worth, and where momentum is pointing.

It’s a bit like surfing. The fundamental analysis tells you whether a wave is worth catching. The technical analysis tells you whether the wave is building or breaking.

Some signals we look at:

Moving averages — Is the stock trending up or down over the past 50 or 200 days? When the short-term average crosses above the long-term average, that’s often a bullish sign.

RSI (Relative Strength Index) — A 0–100 score. Above 70 often means a stock is overbought (too much enthusiasm). Below 30 often means it’s oversold (too much fear). Around 60 is often a healthy, upward-trending stock.

MACD — A momentum indicator that shows whether buying pressure is increasing or decreasing.

None of these predict the future. But they describe what the crowd is currently doing — and crowds matter in markets.

Pillar 3: The Story (Qualitative Analysis)

This is the hardest pillar to automate, and the most interesting.

Numbers only tell you what happened. They don’t tell you why, or what’s coming next.

The qualitative layer tries to answer questions like:

Do the people running this company actually know what they’re doing?

Did management give confident guidance on the last earnings call, or were they vague and evasive?

Is this company in a growing market, or a shrinking one?

Are insiders buying their own stock — or quietly selling?

This is where AI reasoning becomes genuinely useful. A language model can read through an earnings call transcript and pick up on things the numbers miss — the tone of management, whether they raised or lowered guidance, whether their explanation for a bad quarter sounded like a real plan or a cover story. Also it can easily compare...