Q-Alpha

Why Forecasts Should Be Shown as a Range, Not a Point Estimate

Why a single number hides dispersion, how the mean and the median part ways in skewed outcomes, and what removing assumed drift does to a picture of spread. Statistical literacy from Q-Alpha, a research tool, not financial advice.

The short answer

A forecast written as one number answers a question nobody actually asked. People want to know what might happen and how sure anyone can be. A single figure answers only the first half, and it does so with a confidence the underlying evidence rarely supports. A range, or better a whole distribution, puts the uncertainty on the page next to the estimate.

This page explains why that matters, where the average goes wrong when outcomes are lopsided, and why leaving out an assumed direction is the modest way to draw spread. It closes with one panel in Q-Alpha that follows these ideas. Nothing here is financial advice.

What a single number hides

Statisticians call the single figure a point estimate. Its counterpart is the interval or range estimate. The point is a summary of the centre. The range describes the scatter around that centre, and the scatter is where most of the practical information lives.

Consider two situations with the same central figure. In the first, nearly every plausible outcome sits close to it. In the second, outcomes are scattered widely on both sides, and the central figure itself is quite unlikely to be the one that occurs. Quoted as points, the two look identical. Shown as ranges, nobody would confuse them.

A point also invites a habit of mind that a range resists. Once a number is written down, readers anchor on it and treat any other result as a surprise or an error. A range sets the expectation that the eventual outcome will land somewhere inside a band, and that landing away from the centre is ordinary.

Point estimate, confidence interval, prediction interval

Three terms are often mixed up.

  • A point estimate is a single value offered for an unknown quantity.
  • A confidence interval describes uncertainty about a parameter, such as a long-run average. It says how precisely the average has been pinned down, not how far an individual outcome may fall from it.
  • A prediction interval describes where a single future observation may fall. It is wider than a confidence interval for the same data, because one observation varies more than an average does.

Reading a narrow confidence interval as if it were a prediction interval is a common mistake. It makes the future look more settled than the method claims. Every interval also rests on a model. If the assumptions behind it are wrong, the band is wrong too, and it is usually wrong in the direction of being too narrow.

The mean-versus-median trap in skewed outcomes

When outcomes are symmetric, the mean and the median sit close together and either one is a fair description of the centre. Many real quantities are not symmetric. Incomes, project overruns, insurance losses and compounded growth over long periods all have a long tail on one side.

In a right-skewed distribution a few very large outcomes drag the mean upward. The median, which only marks the middle of the ordered list, hardly moves. The result is an average that most individual cases fall below. Quoting that average as what to expect describes an outcome that is better than the typical one, without any dishonesty in the arithmetic.

This is why a careful summary of a skewed quantity reports the median, or both figures, and shows the shape. If someone gives you only an average for something with a long tail, the first question to ask is where the median sits.

Why removing assumed drift is the modest choice

Any picture of future spread has two ingredients: a direction and a width. The direction, often called drift, is the assumed tendency to move up or down. The width is the dispersion.

The two are not equally knowable. Dispersion can be described reasonably well from past behaviour, although it changes over time. Drift is far harder to estimate, and small changes in the assumed drift move the whole picture. A chart that bakes in an upward drift is making a claim about direction, whether or not it says so, and viewers tend to read the tilt as a promise.

Setting drift to zero removes that claim. What remains is a picture of how widely outcomes might scatter, centred on no change. It does not say that no change is expected. It says the picture takes no position on direction and shows width only. That is a smaller claim, and it is one the available evidence can actually carry.

How Q-Alpha draws this

Q-Alpha is a web application for market and portfolio research. On its Risk Index screen, below a result, the app renders a panel titled driftless scenario distribution. The panel is deliberately limited:

  • It assumes no upward or downward drift. The spread is centred on zero.
  • Its width is derived from the result's own volatility score, so a calmer result draws a narrower shape and a more volatile one draws a wider shape.
  • It is generated deterministically for each symbol. The same result draws the same picture each time instead of flickering between random versions.
  • It is labelled an illustration, not a prediction, and carries short reminders that it is not a target, not a direction and not advice.
  • It marks both the mean and the median and includes a note on why an average can drift away from the typical case.

Two limits should be stated plainly. First, the shape is an illustration built on the screen from summary scores. It is not a record of simulated market paths, and no point or band on it is an expected outcome. Second, the illustration is drawn symmetrically around zero, so on the panel itself the mean and the median sit close together. The note about the trap describes the general lesson for skewed outcomes; the panel does not demonstrate skew.

For what the rest of the result shows, including the breakdown and the data-source label, see what a single risk number can and cannot tell you. For scoring a series of your own, see preparing your own price and volume data.

Reading any range sensibly

  • Ask what the band is a band of: an average, or a single future outcome.
  • Ask what assumptions produced it, and whether a direction was assumed.
  • Look for the median as well as the mean whenever the shape is lopsided.
  • Treat the edges of a band as soft. Outcomes beyond them are uncommon under the model, not impossible.
  • Remember that a range describes the model's uncertainty. It cannot include what the model does not know about.

Access

The panel appears under a Risk Index result in the Q-Alpha dashboard. You can create an account, and the pricing page lists what each plan includes.

Q-Alpha does not provide financial, investment, tax or legal advice, and nothing on this page is a recommendation to buy, sell or hold any asset.

Create a Q-Alpha account to see the driftless scenario distribution under a Risk Index result. It is an illustration of spread for research, not a prediction and not financial advice.

Get started