Stop Trying to Predict the Market — Your Research Notes Should Track Decisions, Not Forecasts
Open almost any retail investor's research notes and you'll find the same thing: a watchlist, a fistful of price targets, and a paragraph arguing why a stock is "going to" do something. The whole apparatus is pointed at the future — at prediction. By the end of this piece you'll be able to re-point it at the one thing you can actually control and improve: the quality of your decisions.
That sounds like a semantic dodge. It isn't. It changes what you write down, what you review, and what gets better over time.
TL;DR - The consensus says good research = better forecasts. The evidence says forecasts are mostly noise, and accurate ones rarely beat a low-cost index. - You can't get feedback on a prediction you never scored — and almost no one scores theirs. - What you can improve is your process: the decisions, their reasons, and the conditions under which you'd be wrong. - Track decisions, not forecasts. Forecasts are an input to a decision, not the artifact worth keeping. - Practical version: a decision journal, reviewed on a schedule, beats a watchlist of price targets.
The consensus is reasonable — let's steelman it first
The mainstream view goes like this: investing is about predicting the future value of an asset, so better research produces better predictions, and better predictions produce better returns. Read the 10-K, model the cash flows, form a view, set a target. If your view is more accurate than the market's, you profit. This is not a strawman — it's the literal logic of fundamental analysis, and it has made some people very rich.
It's also internally consistent and pedagogically clean. It gives a beginner something concrete to do. And in narrow, information-rich niches, diligent analysts genuinely do find mispricings. I'm not going to pretend none of that is real.
The counter-thesis: prediction is the part you can't improve
Here's where it breaks. The consensus quietly assumes you can get better at forecasting through practice. For the kind of forecasts most investors make — "where will this stock be in a year" — that assumption is poorly supported.
| What the consensus assumes | What actually holds up |
|---|---|
| Research sharpens forecasts | Most security forecasts are barely better than chance over a year horizon |
| Accurate forecasts → market-beating returns | The majority of active funds — staffed by professional forecasters — trail their benchmark over 10+ years |
| You learn from being right or wrong | You learn nothing if you never recorded the prediction and its reasoning |
Philip Tetlock's two decades of forecasting research found that expert predictions in complex domains were, on average, only marginally better than crude baselines — and that confidence and media presence were negatively correlated with accuracy. The S&P SPIVA scorecards, year after year, show most professional active managers underperforming their benchmarks over long horizons. These are people who forecast for a living, with better data than you have. If prediction skill were the lever, they'd be pulling it.
The deeper problem is the feedback loop. To get better at anything you need a tight cycle: attempt → outcome → correction. A price target you never write down, with reasons you never record, that you quietly forget when the stock moves — that loop is broken at every joint. You can't grade what you didn't log, and you can't improve what you can't grade.
What you can actually improve: the decision
You can't reliably make your forecasts more accurate. You can make your decisions more deliberate, more consistent, and more reviewable. The shift is to treat a forecast as a disposable input and the decision as the artifact worth saving.
A decision is recordable, gradeable, and improvable in a way a vibe about a stock isn't:
The decision record — five fields 1. The decision — what you did (bought, sold, passed, sized up). 2. The reasons — the 2–3 things that actually drove it. 3. The disconfirmer — what would have to be true for this to be wrong. 4. Your confidence — a number, so you can check calibration later. 5. The state of the world — what you knew and felt at the time.
Notice what this captures that a price target doesn't: the reasoning and the disconfirmer. When you review it months later, you're not asking "was I right?" (mostly luck) but "was my reasoning sound given what I knew?" (skill, and improvable). That second question is the only one with a useful answer.
A worked example
Say you're looking at a company after a rough quarter. Two ways to write it up:
Forecast-first (the trap): "Beaten down on a temporary supply issue. Fair value ~$80, 40% upside. Target: $80 in 12 months." Twelve months later it's $62. Were you wrong? About the price, yes. About anything you can learn from? Unknown — you didn't write down why $80, or what would prove the thesis broken.
Decision-first: "Bought a half position. Reasons: (1) the supply issue looks one-off per the call transcript, (2) balance sheet survives 18 months of this, (3) insiders bought. I'm wrong if the issue recurs next quarter or gross margin stays below 30%. Confidence: 6/10. Note: I'm anchored on the pre-drop price." Twelve months later, whatever the price, you can audit it: did the supply issue recur? Was the margin call right? Was 6/10 honest? Now you're learning.
"But predictions are unavoidable" — the honest caveat
To be fair to the other side: every decision contains an implicit forecast. Buying a stock is a bet it'll do better than the alternative. You can't escape prediction entirely, and I'm not claiming forecasts are worthless — a forecast you can defend is better than a hunch.
The claim is narrower and, I think, harder to dodge: the forecast is not the thing to optimize or archive. Make your best estimate, fold it into the decision as one input among several, write down the decision and its logic — then let the prediction go. Optimizing the prediction is optimizing the part you can't control at the expense of the part you can.
Common mistakes when you make the switch
| Mistake | Why it bites | Fix |
|---|---|---|
| Logging the decision but not the disconfirmer | You can rationalize any outcome after the fact | Always write what would prove you wrong, before you know |
| Recording confidence as words ("pretty sure") | Words aren't gradeable for calibration | Use a number; review whether your 7/10s come true ~70% of the time |
| Reviewing only your losers | Right-for-wrong-reasons is just as dangerous as wrong | Review winners too — was the reasoning sound? |
| Editing the note after the outcome | Hindsight quietly rewrites what you "knew" | Append-only; never touch the original entry |
| Reviewing ad hoc | The feedback loop only closes if it actually closes | Put it on a schedule — monthly or quarterly |
Summary + next step
The consensus points your research at prediction. The evidence says prediction is the part you can't reliably improve — and that the feedback loop you'd need to try is almost always broken. So re-point your notes at the decision: what you did, why, what would make it wrong, and how sure you were. That's recordable, gradeable, and improvable. Forecasts are an input you spend and discard; decisions are the artifact you keep.
The concrete tool for this is a decision journal — see [The Investing Decision Journal](the-investing-decision-journal.md) for the format and the review cadence. And when you do form a view worth defending, write it up as a falsifiable [investment thesis](what-is-an-investment-thesis.md) with its disconfirmers attached, so future-you can grade the reasoning, not just the price.