It is easy to criticise investment forecasting. I do it myself on a regular basis. As enjoyable a pastime as this is, there is a problem. All investors are consistently making forecasts. Consciously and unconsciously. Even the biggest sceptic of the folly of market and economic predictions is inevitably expressing views about the future when they make an investment decision. If we are all forecasters, then we need to be careful about denouncing it in a broad and unequivocal fashion. Some nuance is required. If we must make forecasts, when and how should we make them?
If a twenty-year-old has forty years until retirement, most people would agree that investing all or nearly all of their pension across global equity markets is likely to be a sensible decision. Although it doesn’t automatically feel like one, this is a forecast. We are predicting that equity returns are likely to outstrip other major asset classes over the long-run horizon. I don’t like making forecasts about financial markets, but I feel comfortable with this one.
To understand why this is, we need to distinguish between the characteristics of different types of forecasts. Below are four forecasts and next to them is my level of confidence in each:
– Global equity returns positive over next forty years: 99%
– Lower returns from 60/40 portfolio over next decade than the previous decade: 90%
– Non-US equity returns higher than US equity returns over next decade: 65%
– Non-US equity returns higher than US equity returns over next 12 months: 50%
These are all forecasts but my conviction in them varies wildly, from being as close to certain as I would be comfortable, to the toss of a coin. What are the features that differentiate them?
Time horizon: Although it instinctively feels easier to make a forecast about tomorrow than the distant future (much more will have changed in the case of the latter) this is not true for certain types of investment forecasts. Would you rather take a bet about whether equities were higher or lower tomorrow, or over the next twenty years? In many cases extending the time horizon means that the outcomes are less driven by noise and randomness, and more by the fundamentally important variables. For equity markets over the long run this is real earnings growth, inflation, and dividends (to varying degrees) rather than sentiment or flow. Extending the time horizon is only helpful if we are confident what the critical variables are, however.
Critical variables: When we are making an investment forecast; we are usually predicting the behaviour of a host of other variables. We need to correctly gauge what these are and what will happen to them over the forecast period. This is why most investment forecasts are so difficult and pointless. Even if we identify what the meaningful variables are (a huge feat in itself) we might not accurately judge how they will change or influence other variables. The reason I have no idea about whether non-US equity returns will be higher than US equity returns over the next 12 months is because over such a short period I don’t know what the determining factors will be and, even if I did, I still would not be able to anticipate how they might behave. What matters to equity markets over short time periods can vary wildly from year to year; it might be elections, central bank activity or an event I don’t expect to occur, such as a pandemic. Yet even if I knew what mattered in advance in most cases I would still need to foresee the outcome and (if I am predicting price movements) how other investors would react to it. In most cases, it is just too difficult.
Prior knowledge: The investment forecasts that we should be most comfortable making are those where we understand the variables that will drive the outcomes (often by extending the time horizon), and where we don’t also have to predict the level of those variables, because we already know them. Why do I have a high level of confidence that a 60/40 portfolio will produce lower returns in the next decade than the last? Because over a ten-year period, the critical driver of the performance of a 60/40 portfolio is likely to be the starting valuations. I know that yields are far lower and equites more expensive (I can observe this, I don’t need to foretell it) so my conviction in this can be strong. That doesn’t mean I am certain, there are scenarios where this forecast doesn’t come to pass but the chances of these seem remote.
Conviction: In most cases our investment forecasts will reside somewhere between the near certainty of equities producing a positive return over the very long-run and the near-randomness of what they will do over the next quarter. Our views therefore need to reflect this. My 65% confidence that non-US equities will outperform US equites over the next decade is because over such a time horizon there is a strong relationship between starting valuations and subsequent returns. The conviction I hold in this view, however, is tempered by three factors: 1) Although the relationship between valuations and ten-year returns has tended to work historically it has not always – it is not an unimpeachable association. 2) Other factors may also matter over this time horizon (earnings growth / return on equity / sector composition); therefore, I cannot be confident that other variables won’t be more influential. 3) The relationship between starting valuations and subsequent ten-year returns may be an artefact of history, perhaps it no longer holds.
When we are making a forecast it pays to have a checklist to ensure we know exactly what we are forecasting and whether we should be:
– Do I know what I am forecasting?
Although the headline forecast is obvious, the result will be driven by other factors. We must be clear about what these variables are and what we are really predicting. When we are forecasting the performance of asset classes or securities over the short-term, we are attempting to anticipate investor behaviour.
– Am I also predicting the level of the influential variables, or do I know them already?
If I am making a prediction about where ten-year government bond yields will be in a year’s time, I might be making assumptions about inflation expectations and the elements that influence this. It is rarely just a single forecast. Forecasts are more robust if the variables are known (valuations) or easier to foresee (economies will grow).
– How much will randomness and noise impact the outcome?
The more randomness and noise there is in an outcome the less sensible it is to make forecasts. Have forecasts in this area worked in this past for us or others?
– How does the time horizon impact the forecast?
Time horizon is critical and materially impacts the amount of luck involved and what variables might be at play. For financial markets, short-term forecasts are usually incredibly problematic.
– Is there a historically strong relationship between those variables and the outcomes I am predicting?
Although not a failsafe, having some evidence of a robust relationship between the thing we are forecasting and the variables that impact it is critical.
– Have we considered what breaks the relationship?
If we are basing a forecast on historical relationships between different variables, we must assess what might break them.
– Have I expressed a level of conviction?
We should always express a level of confidence about any forecast or prediction we make. Not only does it force us to be explicit about the uncertainty in the opinion we hold, but it makes it far easier to change our mind or adjust our view.
– Have I stated what will make me change my mind?
To make changing our mind less painful in the future it pays to state at the outset what developments might cause us to alter our perspective.
– How much specific risk am I taking?
The more idiosyncratic risk in a forecast, the more dangerous it is because the number of potentially impactful variables expands dramatically. I am extremely confident that equity markets will make positive returns over forty years, I would not be confident about making such a claim for any single company over this period.
Investors should make as few forecasts as possible and avoid at all costs making narrow, specific or short-term forecasts, where our success rate will be miserably low. As much as we might like to, however, we cannot avoid making a certain number of forecasts. When we do, we must understand exactly what it is we are predicting and ensure that the evidence is firmly on our side.