Refusal to Forecast
“The odds of consistently and correctly forecasting Federal Reserve decisions, GDP figures, Brexit, Trump or quarterly earnings are vanishingly small, evanescent and deeply competitive.”
The principled rejection of precise short-term predictions about stock prices, earnings, or macroeconomic outcomes — grounded in a Popperian epistemology that treats such forecasting as pseudo-science in a complex adaptive system.
Anderson drew explicitly on Karl Popper's philosophy of science: the claim that any system is 'falsifiable' requires that it make precise predictions. But financial markets are complex adaptive systems where participants respond to predictions, rendering prediction inherently self-defeating. Anderson's response was not agnosticism but a different form of conviction: rather than predicting near-term outcomes, he sought companies with the widest possible range of positive potential outcomes and held them long enough that the distribution of outcomes could work in his favor. This is distinct from dismissing analysis — Anderson was deeply analytical about business quality, management capability, and market structure. He simply refused to convert that analysis into the pseudo-precision of a price target.
Refusal to Forecast
Definition & Origins
The refusal to forecast is Anderson's principled rejection of precise short-term predictions — of earnings, cash flows, share prices, or macroeconomic outcomes — grounded in the conviction that such forecasting is pseudo-science when applied to complex adaptive systems. It is not agnosticism and not laziness: it is the deliberate substitution of scenario-based, probability-weighted thinking for the industry's ritual of point estimates.
The intellectual roots are explicit. From Karl Popper comes the epistemology: knowledge advances by conjecture and refutation, and any claim to certain foresight in an open system is a category error. From the Santa Fe Institute's complexity science comes the mechanics: markets are adaptive systems whose participants react to predictions, rendering the predictions self-defeating. And from personal trauma comes the conviction: "Personally I lost any belief in the twin notions of predictability and efficiency on October 19, 1987. The S&P 500 losing 20 per cent of its value on no news at all seemed a little hard to rationalise away" (Graham or Growth).
The most quoted formulation appears in Aberration or Premonition?: "We now need a dogged refusal to make forecasts of earnings, cash flows or share prices." Anderson immediately supplies the reason: a forecaster anchored to the most likely near-term outcome "will be most unlikely to hang your spreadsheet on predicting a discontinuity" — and discontinuity is where all the returns are.
Core Ideas
Forecasting destroys exactly the information that matters. Point estimates force convergence on the modal outcome; but investment returns live in the tails. "If you are merely forecasting the most likely outcome over the next year or two... It's much more sensible to predict a continuation of current business or to follow guidance." The forecast thus structurally excludes the transformative case. "Certainty is an abject temptation. The world is too complex, too erratic and too full of surprises to make spot forecasts of anything of significance."
Being "correct" is the enemy of good investing. The most counterintuitive claim: "trying to be 'correct' is the enemy of good investing." The forecaster who is right about next quarter's earnings has purchased nothing of durable value; the investor who is roughly right about the possibility of an extreme outcome has purchased everything. "It's much more valuable to have doubt and to make portfolios the beneficiaries of potential Black Swans." Hence the famous conclusion: "the most likely forecast a dozen years ago was clearly that Amazon would fail. That was rational analysis. But it wasn't a very good assessment of the probability adjusted pay-offs."
Some things are predictable — and they are the valuable ones. The refusal is selective, and this is its most misunderstood subtlety. Anderson distinguishes unknowable short-term events (Fed decisions, GDP prints, Brexit, quarterly earnings) from high-probability long-run processes (Moore's Law, solar cost declines of 15–20% per annum, battery learning curves). "If we know what the underlying improvement rate of a technology is likely to be then the room for debate is minor, the impact is long-lasting and the opportunities are potentially dramatic." The Tetlock-style project of marginally improving forecasts of the unknowable is abandoned; research is redirected to "the predictable and exponentially profitable categories."
Replace the point estimate with the weighted scenario. The operational alternative is spelled out in both 2018 essays: outline "a set of possibilities and probabilities, endeavour to make them extreme, blend them with each other and then think about the potential returns. Then we watch. It's better than acting." The portfolio itself becomes the expression of the view — in his 2022 formulation, "our portfolio is not a prediction. It is a collection of companies where we believe the range of possible outcomes includes genuinely transformative success."
Practical Application
The Roche/Illumina moment. Anderson's favorite illustration of the two methods colliding. Asked how Roche arrived at its $44.50 bid for Illumina in 2012–13, the "young, eager and highly financially trained" CFO answered that he plugged five years of analyst forecasts into a DCF model. At that very moment, Illumina's owners were discussing a business whose extraordinary growth "would only fully blossom well after five years had elapsed." One side priced five forecast years; the other side held the genomic revolution.
Refusing the Tesla price target. The cleanest case of the refusal in public. In Stay on the Road Less Travelled: "it seems implausible that we can estimate either the likelihood of success in a radically new endeavour nor the precise outcomes in cash-flows should success emerge. To us it is bizarre that brokers, hedge fund mavens and commentators can claim to be able to decipher the future and assign a precise numerical target to the value of Tesla. Perhaps they are all geniuses. We are not." The substitute: hold the conviction that EVs would win (a high-probability process), admit ignorance of autonomous-driving economics (an unknowable), and size the position to the asymmetry.
Screening out the noise. The Core Investment Beliefs institutionalize the refusal: "We are very dubious about the value of routine information. We have little confidence in quarterly earnings and none in the views of investment banks. We try to screen out rather than incorporate their noise." The 2022 interview adds the acid test: "even if I knew what the Fed was about to do, I'm not sure it would help me make good decisions."
The Moore's Law thought experiment. Anderson's proof that the refusal is cheap to practice: "if I had just known in — when I started in 1983 that Moore's law was going to continue for the rest of my career at Baillie Gifford, did I need to know anything else?" One durable, high-probability process, taken seriously for four decades, outweighed the entire daily production of the forecasting industry.
Common Misconceptions
Misconception 1: Refusing forecasts means refusing analysis. Anderson was deeply analytical — about battery learning rates, sequencing cost curves, addressable markets, and founder quality. What he refused was the conversion of analysis into false precision. The distinction: analyze the process (knowable, improvable), never pretend to predict the event (unknowable, unpriceable by you).
Misconception 2: It means having no view. The refusal produces strong views — that EVs would beat combustion, that solar costs would keep falling, that platform economics would concentrate. These are convictions about direction and distribution, arrived at with "75 percent confidence" and rising to "somewhere around 90 percent." What is refused is the attached timetable and price target: "It's rare for us to know when a dramatic change will occur but frequent for it to be close to inevitable at some point."
Misconception 3: Humility about forecasting is a pose to deflect accountability. The opposite: the refusal raises the accountability bar, because it removes the alibi of a wrong-but-reasonable model. Anderson's own scoreboard is inspected publicly — the Apple sale ("there is a fallibility about this process"), the Amazon trims ("misguided"), TAL ("a very bad, very serious mistake"). Epistemic modesty about the future coexists with brutal candor about the past — the signature of epistemic humility rather than its absence.
Anderson's Own Words
"We now need a dogged refusal to make forecasts of earnings, cash flows or share prices... Certainty is an abject temptation. The world is too complex, too erratic and too full of surprises to make spot forecasts of anything of significance. I'd push this further: trying to be 'correct' is the enemy of good investing."
"We need to give up the excessive arrogance implicit in forecasts if we are to maximise returns. After all the most likely forecast a dozen years ago was clearly that Amazon would fail. That was rational analysis. But it wasn't a very good assessment of the probability adjusted pay-offs."
"We should respect and endure uncertainty, try to identify where extreme upside might occur and observe patiently."
"The odds of consistently and correctly forecasting Federal Reserve decisions, GDP figures, Brexit, Trump or quarterly earnings are vanishingly small, evanescent and deeply competitive."
Thought Evolution
Key Letters / Related Concepts
Key letters: Aberration or Premonition? (2018) · Graham or Growth (2018) · Stay on the Road Less Travelled (2021) · Scottish Mortgage Annual Report 2017 · Masters in Business Interview (2022)
Related concepts: Epistemic Humility · Imagination in Investing · Power Law of Returns · Long-Termism · Benchmark Irrelevance