What to forecast, how, and over what horizon

What can an analyst forecast?

A forecast object is the quantity you project. Four kinds are common, and the right one depends on the information available, efficiency, accuracy, explanatory value and whether the forecast can be verified.

What the analyst forecastschoose on information, efficiency, accuracy, explanatory value and verifiability

• Drivers of statement lines Memorize subscribers and revenue per subscriber; gross margin %
  • Gain explanatory value, often accuracy
  • Cost more inputs to forecast and update
• Individual statement lines Use it lines without clear drivers, or immaterial ones
• Summary measures Memorize free cash flow, earnings per share (EPS), total assets
  • Gain efficiency: fewer variables
  • Cost less transparent, harder to audit
• Ad hoc objects Use it a loss or gain not yet in the accounts (legal case, disaster)

Object	Example (Lumen Ledger, a software maker)	When to use
Drivers	subscribers × revenue per subscriber	when drivers explain the line and are disclosed
Individual financial statement lines	amortization of acquired software	lines without clear drivers, or immaterial ones
Summary measures	free cash flow, earnings per share (EPS)	when stable and predictable, or when disclosure is thin; less transparent
Ad hoc forecast objects	the cost of a pending data-privacy fine	events not yet in the accounts

Forecasting drivers rather than the line itself has the benefit of explanatory value, because a single line may have several drivers moving in different directions. Summary measures win on efficiency (fewer variables) but lose transparency, so the forecast is hard to audit. An ad hoc object matters when you must judge the size and timing of a loss before the issuer books it, for example after a material legal proceeding, a regulatory action, a tax dispute or a natural disaster.

Principles for choosing objects

• Prefer objects that are disclosed regularly, or can be calculated from what is. Then you can compare each quarter’s result with your forecast and correct it. A market-size figure from a one-off consultancy report can inform your view but cannot confirm a forecast in time.
• Do not build more detail than you can verify or update. If gross margin is reported only for the whole company, a forecast by product line cannot be checked, and a miss tells you nothing about which product was wrong.
• Keep the model proportionate. Very complex models need more forecasts and more time to update, often without better accuracy. Focus on the most important drivers and use aggregates where extra detail adds little. If each region has similar economic exposure, forecasting region by region only adds work.

Exam tip If a question asks which object is best, pick the one the company reports regularly. “More detailed” is rarely the right answer.

Which of the four approaches fits?

For any object there are four general approaches, and in practice they are often combined.

Forecast approachapplies to any object; often combined

• Historical results Memorize past is precedent
  • Fits stable, non-cyclical, unchanged structure; immaterial lines
  • Avoid cyclical, restructuring, big acquisition
• Historical base rates and convergence Memorize move toward a peer or industry long-run median
  • Fits young or small firm in an established industry with many peers
  • Avoid new or changing industries, highly cyclical, a dominant leader
• Management guidance Memorize management targets, often a range
  • Fits items management controls and has a record on
  • Avoid macro-driven items; weak or promotional record
• Discretionary forecast Memorize surveys, models, probabilities, analogies
  • Fits cyclical, few comparables, no guidance, fundamental change
  • Avoid when it has no evidence behind it

Historical results: the past is precedent

Memorize

Historical results use past observed values, such as a historical median, as the forecast. It is the easiest approach and perhaps the default. It suits companies whose industry structure is not expected to change and that have low sensitivity to the business cycle, and objects that are immaterial or on which the analyst has no view.

It is a poor fit for cyclical companies, because a future period is likely to sit at a different point in the cycle. An “over the cycle” average can be reasonable for a multi-year view of a cyclical company, but it hides year-to-year volatility, so it is poor for a specific year. It is also poor when the company is restructuring or has made a large acquisition or disposal, which makes history non-comparable.

Historical base rates and convergence

Memorize

Historical base rates and convergence take an industry or peer-group average or median, measured over a long period, as a base rate that the object moves toward over a chosen time frame. The base rate can also be a macro variable such as gross domestic product (GDP) growth. It is not a rote exercise: the analyst chooses the object, the peer sample and the speed of convergence.

It suits established industries with many listed peers (banks, airlines, restaurants, retailers), and smaller or younger firms maturing toward the profile of larger peers. A discount grocer with 40 stores pays more per store for warehousing and trucks than a chain with 900, so its distribution cost ratio can be expected to drift down toward the peer median as the network fills in.

It is poor in new or changing industries, where a base rate is hard to compute or less relevant, and in highly cyclical industries, where a smooth convergence hides the swings. It also fails for an industry leader that dominates the base rate. If one company is 70% of the industry, the industry average is mostly the company’s own history.

Management guidance

Memorize

Management guidance is a target management gives for revenue, earnings, capital expenditure or other measures, for the next quarter, year or longer. It is often a range, updated through the year, and it reflects management’s information advantage and forward-looking view. It can also be suspended when uncertainty is high.

Use it when it is provided and management has a track record you have checked, by comparing past guidance with actual results. Management tends to forecast best what it controls, such as operating expenses and capital expenditure. It has no edge over investors on macro variables such as GDP or commodity prices, so guidance is a weak basis for the revenue of a company that is highly sensitive to the business cycle. Always work out the assumptions inside the guidance and test whether they are plausible. Do not assume the midpoint of a range is management’s true expectation: studies find the upper bound is often closer, because a pessimistic lower bound creates a hurdle that is easy to clear.

The analyst’s discretionary forecast

Memorize

A discretionary forecast covers everything else: surveys, quantitative models, probability distributions and analogies to precedents that differ from peer averages. It is most common for cyclical companies, companies with few comparables or no guidance, and companies facing a fundamental change in competition or regulation. For a first-of-its-kind home battery, for example, an analyst might combine pilot installations, utility tariff data and an installer survey into several adoption cases. It needs evidence: an unsupported view is just a guess.

Go deeper Why complex models rarely forecast better

Each extra input is another forecast that can be wrong, and another number to update every quarter. A model with a hundred product-and-region lines may take months to build and weeks to refresh, and its detailed outputs still cannot be checked, because the company reports only totals. Accuracy rarely improves. A model on a few important drivers that reconciles to disclosed numbers is usually more useful, and it is easier to explain.

Go deeper An over-the-cycle median versus a year-by-year path

For a cyclical firm there are two sensible treatments. A multi-year valuation can use an over-the-cycle median of margin or growth, which is fair for the long run. But a forecast for next year needs a view of where the cycle is now: a median would show the wrong number in both a boom and a slump. That is why discretionary forecasts, based on the analyst’s read of the cycle, are the most common approach for cyclical revenue. And history alone from a peak year is the one choice that is always wrong.

How far ahead should the forecast go?

The forecast horizon is set by four influences.

Forecast horizonhow far ahead the model goes

• Investment strategy Memorize long-term managers 3 to 5 years; short-term managers the next quarters
• Cyclicality Memorize long enough to reach mid-cycle sales and profit
• Company events Memorize time for an acquisition or restructuring to show its effect
• Employer Use it may fix the horizon outright

• Investment strategy. Most strategies state a time frame and average holding period. Long-term managers often focus on three to five years, while short-term managers may focus on the next one or two quarters.
• Cyclicality. The forecast should run long enough for the business to reach an expected mid-cycle level of sales and profitability. If a cyclical company is near its peak, a short horizon never leaves the peak.
• Company events. A recent acquisition or restructuring needs enough time for its benefits, and its costs, to appear in the statements.
• The employer. Some employers set fixed parameters, and the analyst has no choice.

Exam tip Cyclical company near a peak or trough: the horizon must reach mid-cycle. That is the answer to most horizon questions.