There will be much ado about Apple earnings report on April 23. In order to better understand Apple's results, I want to analyze Apple's new approach to reporting its revenue guidance for the next quarter. The analysis involves computer models and Ferraris.
What I want to address, eventually, is Apple's new way of predicting a range of revenue in its quarterly guidance. But, to do that, I want to invoke an analogy.
Earlier this year, I was reading the February 2013 issue of Car and Driver. Starting on page 22 is a report on how various cars, in specific price ranges, performed at the Virginia International Raceway (VIR), a 4.1 mile course. It's a fun article, but what caught my eye was a comment on page 37 about the Ferrari 458. Ferrari has a computer simulation that can predict a lap time given the car and conditions, and in this event, the Ferrari team told C&D that a "good" lap time would be 2:49.0. (2 minutes, 49 seconds.)
I haven't checked with Ferrari, but from my own computer modeling experience, I'm sure that their digital model has inputs for the weather, elevation, course geometry, road surface and slant angles. And then there are the car parameters: horsepower and torque curves, tire size and adhesion, the suspension and so on. By integrating the forces on the car as it moves around the track, one can predict the lap time.
Image Credit: Ferrari
C&D's driver ended up with a 2:49.9, just a tad off the predicted "good" time. What that tells me is that 1) the magazine's driver almost fully exploited the capability of the car as well as 2) that the Ferrari team has a pretty good model, which they've probably tuned after many years of racing.
Of course, there are uncertainties in the data inputs to the model and there will be approximations to the details of the physics involved. Ferrari could have given C&D bracketed times, but for simplicity, the target time of 2:49.0 was good enough.
Even so, and this is important, lap times by experienced race car drivers vary by a few tenths of a second (if nothing has changed) not by five or 10 seconds. So if the same driver were to rerun the VIR, we'd expect to see, say, 2:49.5 or maybe 2:48.8. But not 3:05.0 or 2:40.0.
Apple's Historical Method
Turning to Apple now. Each quarter, Apple provides guidance to analysts and investors for its revenue in the coming quarter. It used to be that Apple would, like Ferrari above, provide a single number.
The historical problem with a single number, mathematically, is that it's very likely Apple would miss the exact number. Anything less is perceived as a failure. And in the racing example above, we know that there's an expected range given the driver, car and course conditions.
The Scotty Principle
At the last Apple earnings report, Apple's CFO Peter Oppenheimer announced that guidance for earnings would be reported differently from now on. Instead of a single number, Apple would predict a range. You can read more details about that in Jeff Gamet's terrific article from January 23, "Apple Dumps 'Scotty Principle' for Financial Guidance."
Basically, to avoid a misperception problem, Apple would, in the past, provide a conservative estimate, sandbagging, and the company would always exceed that conservative number, making them look very good indeed.
In time, analysts learned that Apple trick. Of course, the downside of the Scotty Principle was that no one could confidently predict how much better Apple might have done. External models were not likely to be as good as Apple's because Apple has access to all kinds of internal sales and customer data. Plus, there was no transparency into Apple's model and its predicted limits. In the end, analysts had enough of that game and so did Apple.
The key section in Mr. Oppenheimer's January 23 announcement was:
To further increase transparency we're changing our guidance approach. In the past, we've given a conservative estimate. Going forward, we plan to project what we are likely to achieve."
Those who are familiar and modeling and statistics will be keen on the word "likely." Another way of saying this is that it is unlikely that Apple's revenue will fall outside the predicted range. For Q2, 2013, to be reported on April 23, that range is [$41 billion - $43 billion].
Expectations of a Model
Any good model has a way of predicting a range of outcomes, based on the uncertainty of the inputs. In the combat models I've worked with that's called a stochastic process. A random number generator is used to vary uncertain inputs, the model crunches and the result is some confidence in a range of likely outcomes.
When Mr. Oppenheimer provides a range, and says he's confident that that represents the likely result, he's not just guessing. Instead, he has some confidence in the range based on expected variations in the input parameters. The better the model is, the narrower the range in expected results.
Because the range of the predicted results, $2 billion, is small compared to the total number, the model would appear to be fairly good.
Analyzing the Results
If Apple's revenue lands inside the $41 to $43 billion range, that's a success. Because the model has an inherent range of likely results, one cannot claim that if Apple comes in at $41.1 billion that that's a failure because it's in the "low" end of the range. Instead, it's a success because Apple performed according to expectations, within the limits of predicted error.
It's like the race course analogy above. The Ferrari team prediction model is analogous to Apple's financial model. The driver's performance in lap time is like Apple's revenue performance. There's a range of expected times, centered on a "good" lap time, and statistics and human error result in a slight variation, by a few tenths of a second. Really bad lap times, a spin out, would be reflected in much bigger variations than the small statistical fluctuations. To put it another way, no other driver (Stig?) who achieved a 2:49.7 would seriously claim that he's a better driver.
Similarly, a "bad" result for Apple can only be declared if the revenue falls below the predicted number of $41 billion. That's an unlikely, unforeseen result. Something changed. Customers dramatically, quickly changed their behavior. A factory went on strike and parts weren't available. Something.
Similarly, anything greater than $43 billion, while a plus, is also unlikely. It would therefore be irresponsible to claim that Apple should have done better. It would be like, to put a dramatic point on it, like a fan at a car race, like the one cited above, saying:
"Well, that magazine's driver is an idiot. He should have been able to turn in a 2:30 lap."
Everyone, the magazine editors and the Ferrari team, would just shake their heads and declare the guy, with his beer can in hand, an uninformed idiot. They understand their model.
Summary of the Analysis
- If Apple comes in at less that $41.0 billion, that's failure. Tim Cook will have to explain.
- If Apple comes in between 41.0 and 43.0 billion, that's a success, without qualifications. Quibbling inside the predicted range is mathematically irresponsible analysis.
- If Apple comes in at greater than 43.0 billion, that's also a success. But the number is exceptional, and was not foreseen by the model. Nor could anyone have foreseen it.
Even so, I predict that if Apple's revenue lands in the low end of the predicted range, untrained observers will be quick to declare Apple's failure. It will be sad to watch.