Best Practices Equity Research Analysts

What Actually Separates Decent Analysts From the Rest

Most equity research comes out about as useful as a screen door on a submarine. I've spent roughly twelve years doing this work across a few different buy-side and sell-side outfits, and the pattern is always the same. The analysts who survive long-term aren't the ones with the fanciest models. They're the ones who avoid the mistakes everyone else makes repeatedly. The term Best Practices Equity Research Analysts shows up in job postings and internal memos, but it rarely means anything specific. Here's what it actually looks like when you strip away the corporate language.

A Model Is Not a Thesis

This is the mistake that costs people their jobs. A spreadsheet with thirty tabs and a water-fall that changes color when you toggle an assumption is not research. It's an accounting exercise dressed up as analysis. The actual work starts when you can state your investment conviction in one sentence without any numbers attached to it. I had a boss once who made us write a single paragraph thesis before building a single cell in Excel. If you couldn't explain why a stock was mispriced without referencing revenue multiples or terminal values, you didn't have a thesis. You had a calculation looking for a question. This practice cuts model-building time from two days down to about four hours for most names, because you stop adding features you don't actually need.

Source Your Assumptions Or Don't Write Them Down

Every assumption in your model needs a traceable origin. Management guidance, competitor filings, industry data from a paid source, your own bottom-up build from first principles. When you can't point to where a number came from, it's not an assumption. It's a guess wearing a tie. I learned this the hard way during a coverage change in 2018. I was covering a mid-cap industrial company and had built a detailed margin expansion story based on management commentary from two earnings calls. The model looked clean. The stock went from twenty-two to forty-three over eighteen months. Then I went back to verify my source material for a client presentation and realized I had misread a single qualifier in the Q2 call transcript. Management had said margin expansion was conditional on raw material costs staying below a specific threshold. Raw material costs did not stay below that threshold. The stock dropped forty percent in three weeks. I had no documented source that captured the conditionality because I hadn't written it down at the time. After that, every assumption gets a footnote with the exact source, date, and link. It adds about twenty minutes per model but it's the difference between looking like an analyst and looking like someone who winged it.

What The Process Actually Looks Like Day to Day

Start with the financial statements. Not the press release, not the summary on Seeking Alpha, the actual SEC filings. 10-K, 10-Q, 8-K when something material happens. Read the notes to the financials. That's where companies hide the things they don't want you to notice. Lease obligations, pension assumptions, revenue recognition policies, segment restructurings. The income statement tells you what happened. The notes tell you why it happened and whether it will happen again. Then read the earnings call transcripts. Not the prepared remarks. The Q and A section. That's where management either reveals something useful or reveals how little they actually know. I track questions that get deflected or answered with vague language across multiple quarters. If the CFO keeps saying "we're working on it" about gross margin without ever giving a timeline, that's data. Write it down. Build the model after you understand the business, not before. Start with revenue drivers. What actually drives top line growth for this company? Unit volume, pricing power, acquisition contribution, foreign exchange? Get that right and the rest follows. Get it wrong and you're optimizing the wrong lever.

Discount rates are where most models go to die. WACC isn't a number you copy from a template. It changes with capital structure, with market conditions, with the cost of debt at current rates. I've seen analysts use a fifteen percent WACC for a utility and a seven percent WACC for a high-growth software company. The software company was correct. The utility should have been closer to eight. Using industry averages is faster but it's wrong about half the time in my experience.

The DCF Trap Nobody Talks About

Discounted cash flow models are taught in every finance program and they're almost useless for practical research unless you know what you're actually measuring. A DCF doesn't tell you what a stock is worth. It tells you what assumptions you need to believe for the current price to be fair. That's a different thing entirely. When I run a DCF, I back-solve it first. What growth rate and what terminal multiple does the market imply at the current price? Then I ask whether those implied assumptions are reasonable given what I know about the business. This takes about ten minutes and it's more useful than three hours of forward projection work. Most companies don't grow perpetually at five percent. Most don't trade at five times free cash flow forever either. The back-solve tells you what the market is actually pricing in.

Write Like You're Talking to Someone

Research reports should be readable by a competent person who has never seen the company before. If you need three paragraphs of jargon to explain why you think a stock will go up, you probably don't know why either. I write first drafts as if I'm explaining the investment case to a colleague over coffee. Then I edit for clarity and remove the filler. This process usually takes about an hour for a standard report and the result is something people actually finish reading. The recommendation matters less than the reasoning. "Buy" with no clear catalyst is noise. "Hold" with a detailed risk framework is useful. I've seen junior analysts lose credibility by being cheerleaders. The analysts who build long-term reputations are the ones who get credited when they're wrong, because they say exactly what they believe and what would change their mind. That last part is critical. State your falsification conditions explicitly. If the stock hits forty dollars and your thesis is still intact, your thesis is wrong. You need to know what outcome would prove you incorrect before you make the call.

Coverage Universe Management

You cannot maintain high quality across fifty names. The math doesn't work. Most solid research teams cap individual coverage at eight to twelve stocks depending on sector complexity. When you spread thinner than that, you stop doing original research and start summarizing what others have already published. That's not analyst work. That's editorial work. If you're maintaining coverage outside your core competency area, flag it. I once got pulled onto a semiconductor equipment name because the previous analyst quit and nobody wanted it. I covered it for six months. My model accuracy and comment quality dropped noticeably because I didn't have the supply chain relationships or the channel check network for that sector. The right answer was to partner with someone who actually covered that part of the market, not to pretend I could do it alone.

Common Pitfalls That Kill Reports

Copy-pasting competitor financials without adjusting for accounting differences. Revenue recognition varies enough between companies that raw top-line comparisons can be misleading. One company capitalizes certain costs that another expenses. A third might bundle services into product revenue. Adjust or don't compare. Ignoring share count changes. Dilution from options, convertible debt, or active buyback programs changes per-share economics dramatically. I've seen analysts model revenue and earnings correctly and then forget that outstanding shares were dropping twelve percent year over year. The per-share story looked nothing like the total dollar story. Predicting cyclical peaks and troughs with linear models. If you're covering a commodity-adjacent business and your forecast assumes mean reversion happens smoothly, you're wrong. These businesses jump. They gap. Earnings don't trend. They collapse and recover in steps. I use scenario ranges rather than point estimates for cyclical names. A range from bear to base to bull case takes about fifteen minutes to set up and it's infinitely more honest than a single line of projections.

Chasing consensus instead of testing it. The consensus estimate is a starting point, not a target. Your job is to find where it's wrong, or confirm why it's right. If your conclusion matches consensus after three months of work, you haven't done three months of work. You've confirmed you weren't necessary.

Tools and Workflow

Bloomberg terminals are expensive and they're standard for institutional work. FactSet and Capital IQ are reasonable alternatives. But the tool matters less than the habit of keeping an organized source library. I use a simple folder structure organized by company and date with raw filings in one subfolder, my notes in another, and model files in a third. Everything links back to a master spreadsheet that tracks which assumptions are sourced and which are mine. It takes twenty minutes to set up per name and it saves hours when you need to revisit a thesis six months later. Python or Excel, pick one and get good at it. Python handles large datasets and repeatable analysis better. Excel handles ad hoc modeling and quick scenario testing. Most working analysts use both. I spend about forty percent of my time in spreadsheets and sixty percent reading, checking sources, and writing. The modeling is the easy part. Best Practices Equity Research Analysts follow are really just disciplined habits that most people skip because they're boring. Read the primary source. Document your assumptions. Test your own convictions. Write clearly. Stay within your circle of competence. The market rewards people who do the boring work consistently, not the people who find clever shortcuts.