This statistical model predicts the probability Apple will beat the analyst consensus earnings estimate. The chart below shows the historical correlation between the Earnings Beat Indicator (EBI) the model produced and the actual earnings beat percentage that Apple reported. All earnings results since March 2010 are shown. The average, linear correlation is represented by the dashed line. Data points in the red area are considered a “true miss”, where Apple’s actual results deviated significantly from statistical expectations.
Note: As of Apr 2013 earnings, the statistical model has been retrained using data from Jul 2012 forward. There has been a distinct statistical shift in Apple’s earnings patterns since that point. Therefore the chart below should be read with the expectation that results from Apr 2013 onward has the same reliability as the results prior to Jul 2012. Confidence in this model will rise with new each new data set.
The model currently says:
Earnings Beat Indicator for Apr 2013: 0.74
Last updated Apr 23, 2013
What does the model use for inputs?
All inputs come directly from Apple itself, from the data published in its own earnings announcement. Apple’s publication includes the earnings results, consolidated statement of operations, balance sheet, and cash flow. There are some key pieces of data in there that are very predictive. The model then analyzes those data inputs with moving averages, trends, and statistics (means, standard deviations, etc), to arrive at its predictive output. The one input not from Apple is the consensus analyst estimate itself, of course. That’s the earnings expectation against which the model is calculating the probability of a beat.
Does that mean the model can predict Apple’s earnings?
It’s not so much that the model can predict Apple’s earnings, but rather that it can detect when analyst’s earnings estimates deviate from statistical probability. That’s an important distinction. Apple’s earnings vary. Some quarters are better than others. The results can be analyzed to determine what is probable and what is not. Would you expect Apple to report a loss of $100 billion next quarter? No, of course not. We can all probably agree the probability of that is virtually zero. Similarly, it is highly unlikely Apple will report a profit of $100 billion. It’s those values in the middle near historical actual results of the past several years that are more likely. That’s what the model is analyzing, and then predicting the likelihood of a particular outcome. The next earnings release is unlikely to deviate (much) from the recent past. Apple is a very large company with a very large customer base. It is a slow moving consumer eco-system that Apple itself has a very good handle on. Things don’t turn on a dime, and rarely does the profit capability of the company change much within the 3 months that constitutes the wait time until the next earnings announcement.
How do the analyst estimates figure into this?
The analyst consensus estimate is the outcome for which we want to know the probability of beating. The thing is, analyst estimates tend to be inconsistent as a group, with wildly differing opinions among them, many unreliable in accuracy from quarter to quarter. A statistical analysis of the estimates themselves shows that the analyst estimates have a wider degree of variability (standard deviation) than Apple’s actual results.
Why this is true is probably multi-faceted, including being subject to emotions and outside influences, influenced by sentiment, optimism, pessimism, peer pressure, and other factors. I would argue that they also have a tendency to assign importance to all sorts of factors that just don’t matter as much as they think they do: things like the US economy, the world economy, economic crises, currency movements, partner agreements, competitive factors, etc. Sure, they may matter over the long term, but do not appreciably affect the next earnings announcement. The Apple ship is very slow moving with lots of momentum. Its success will not change much in the next 3 months. And that’s as long a timeframe as we’re dealing with here.
The model’s primary puropse is to detect and alert us of an analyst concensus estimate that is out of line with what the statistics support. It will be harder for Apple to beat an estimate that is too aggressive, and easier for it to beat an estimate that is conservative.
What if Apple posts a true miss?
The model will not give advance warning of that. If the probability of a beat is high (high EBI value), and Apple posts a miss, then that’s what we mean by “a true miss”. Remember, Apple’s earnings results have their own amount of variability. The model is designed to be concerned with how Apple’s satistical variation compares with that of the analyst estimates, and reporting that deviation as a probability of an earnings beat based on those statistical averages.
How was the model able to predict the Oct 2011 earnings miss when almost every analyst, pro and independent, had it wrong?
The answer lies in the question itself. Virtually every analyst, pro and independent, did get it wrong. As a group they were too euphoric, affected by excessive optimism, group think, and peer pressure. A statistical analysis of the numbers Apple itself publishes simply didn’t support such optimism for earnings that quarter. It is exactly because everyone was so wrong that the model was able to predict with high confidence that an earnings beat was not very likely, and warn of the much higher probability of an earnings miss. The model gave a clear warning sign in this case. The blue dot in the lower left is the data point for Oct 2011.
Why was the model not able to predict the July 2012 earnings miss?
Apple’s results were significantly out of line with their historical trends and averages. Read more in the post covering the July 2012 Earnings Miss.