Seeking Alpha
, Think Finance (1,462 clicks)
Long/short equity, arbitrage, event-driven, research analyst
Profile| Send Message|
( followers)  

Often, (NASDAQ:AMZN) is promoted, both by professional analysts and neophytes, with the argument that its margins are temporarily depressed by an investment cycle, and as soon as this cycle is over profitability will flourish mightily. This view allows analysts to stick large margin assumptions into growing future revenues and come up with tremendous prospective earnings and, concomitantly, tremendous present values and targets. As proof, the analysts point to 2009/2010, when indeed showed decent profits.

The whole of this theory can be challenged from many perspectives. Starting with the fact that accounting tries to match expenses with the benefits from investment (investments are depreciated over time); continuing with the fact that this theory has been around for 3 years in which time any investment would necessarily already have to be bearing fruit; and finally, from the mere observation that other past profitable growth companies like Wal-Mart (NYSE:WMT) never had to sacrifice their growing profits at the altar of investment, in spite of investing as much as

But even challenging this theory from these many perspectives has not been enough to move the believers. So this is yet another attempt. I will show how's profitability in the past 11 years was predictable and structural. I will do so with a set of assumptions that will remain mostly unchanged during the entire period, yet predict very faithfully the profitability displayed by The profitability measure I'll use will be the one promoted by Consolidated Segment Operating Income (CSOI).

This exercise will prove three points:

  •'s lack of profitability is structural;
  •'s profitability will thus remain low in the next few years;
  • The main driver of's profitability is the sales mix.

The model

Contrary to common knowledge, was not always as secretive as it is today. Up until 2002, actually provided a glimpse of the profitability for each operating segment - not just by geography - it still does that today - but also in terms of media, EGM and services.

From this glimpse, we could see that media was at around 11% CSOI margins versus revenues, EGM was at a negative 11% CSOI margins versus revenues, and services had a nice 17% CSOI margin versus revenues. However, these numbers were for the U.S., the International segment had not yet matured and traded with lower margins. On the other hand, the "services" segment would be gone from disclosure for several years, substituted by an "other" segment in disclosures that was not the same.

This is how's revenues by segment looked like since 2002:

(2013 is estimated by using the distribution of the first 9 months of the year, annualized to show a total that's the same as the current consensus revenue estimate)

The same, in terms of percentage of total revenues:

We can already see that there has been a significant shift in mix over the years. was overwhelmingly a media company back in 2002, with nearly 80% of revenues coming from media. Presently, is more of a generalist EGM retailer, with 64.8% of revenues coming from EGM, more than twice what it makes from media. Lately, "other" revenues, which include AWS and advertising, have been gaining relevance as well.

This change is already problematic. While had shown itself able to run a specialized media retailer at a positive CSOI margin, no such evidence exists for it retailing EGM at a profit. Worse still, the historical evidence which does exist for mail-order generic merchandise retailers is a negative one, with the failure of Montgomery Ward and the Sears catalog. So this is where it starts to get interesting.


The model will be constructed by applying the observed % of revenues for each segment, to a margin assumption for each segment derived from the margins directly observed in 2002.

The only changes will be that we will assume that media saw its profitability stabilize at 11% since 2003, while EGM worked towards eliminating its large deficit, going from -15% of revenues in 2002 to -2% in 2008, where maturity is achieved and other revenue was always around 10% CSOI profitability.

For instance, for 2002 the calculation would have been 78.8% x 10% + 19% x (-15%) + 2.2% x 10% = 5.3%. This compares to actual CSOI margin of 4.6% in 2002.

You will notice that these sets of assumptions produce an entirely stable profitability profile for all segments for the last 6 years, encompassing both's most profitable years and the latest supposed lack of profitability due to "investment." This is what we get.

(click to enlarge)

It looks like there are a few stray observations, but R2 is at 0.7754, already quite high. The historical performance of this model fits with reality as follows.

(click to enlarge)

We can see clearly that there are two large departures from the model, 2006 and 2009/2010. In the first departure the model overstates profitability; in the second departure it understates it.

This is where it gets interesting. Both of these departures can easily be explained!

  • During 2006, launched AWS;
  • During 2009/2010, saw the sales of its Kindle eReader explode at a time when manufacturing the device was still profitable.

The launch of AWS

The impact from the launch of AWS is plain to see in technology costs. These were rather stable except for 2006, where the launch of AWS led to a 0.9% (in terms of revenues) jump concentrated in just that year. In recent years these costs are once again climbing as a percentage of revenues, but here the reason is clearly the fact that puts all the costs of running AWS in this cost line, so as it gains in size it inevitably leads to more technology costs.

Thus, unlike today, the technology costs for launching AWS back in 2006 were really one-off. And they depressed observed profitability compared to the model.

The Kindle

The Kindle eReader went through a different process. Launched in late 2007 at $399, the Kindle met with early success. And its sales then expanded powerfully over the following years, up until 2011 when it started losing ground to LCD-based tablets.

But the interesting thing regarding the Kindle, is that was not always a promoter of the supposed philosophy of launching devices at cost and making money on content. The early Kindles were profitable. Indeed, maybe even very profitable.

In an iSuppli teardown of the Kindle 2 back in 2009, iSuppli put the BOM at $185.49, when the Kindle was still selling for $349. Over 2009 the Kindle 2 would see prices of $349, $299 and $259, but these were all profitable. And even during 2010, the Kindle would see prices of $259, $189 and finally, $139, converging to breakeven at the end of the year (the BOM is sure to have been lowered meanwhile as well, as confirmed by subsequent teardowns of new models).

At this time, the Kindle was seeing a massive increase in sales. Using eInks's revenues, deducting the existing 2006 revenues (because there was no Kindle then) and then applying a 70% share for the Kindle from 2007 onwards, allows us to have a rough estimate of how many Kindle units were being sold (at $90, $60, $45, $35 per display, with $60 and $30 coming from iSuppli teardowns, the others being interpolated). This gives us the following (Source: Digitimes):

(click to enlarge)

(click to enlarge)

We get 492,000 Kindles in 2007, growing to 19.6 million in 2011, and then dropping going forward. These are rough estimates, of course, but they allow us to see that was selling a decent number of units in 2009 (3.34 million) and 2010 (8.19 million), while the Kindle was still profitable to sell.

So what happens if we take these units, and see what kind of profit margin dollars they could have produced in 2007-2010 (from 2011 onwards there was no longer any margin)? This is what we get.

(click to enlarge)

So between the expanding quantities of Kindles sold and their profitability, the Kindles were able to add 0.5%-1.6% to the profitability of a few of's years. For a while there, was just like Apple (NASDAQ:AAPL), making money in the devices. And for that little while, detached itself from the inexorably declining profitability of its underlying business.

The model, revisited

What happens when we then add 0.9% to's profitability in 2006 to account for the AWS launch costs, and remove the excess Kindle profitability from 2007-2010? This is what happens to the model:

(click to enlarge)

And in terms of evolution, this is what we get, a near perfect fit. The remaining disconnect encompasses just the initial years of Kindle selling (where timing for eInk sales, BOM, etc, are very fluid) as well as the 2007/2008/2009 financial crisis, where might have lost some profitability trying to compensate.

(click to enlarge)

The model projections

So what does this model, which fits's past behavior so well, tell us going forward? The first thing to notice is that the fit is so good using what are mostly unchanged assumptions - totally unchanged for the last 6 years.

What this means is that the model is enough to explain the drop in profitability we have observed recently. And it also explains - together with the Kindle anomaly - the years where had its best profitability.

For us to use this model going forward, we need a set of assumptions regarding how's different segments will fare. The set I used was as follows - this is aligned with the growth patterns being observed right now, with EGM growing faster than media, and "Other" services growing faster than either.

Using these quick assumptions, we get the following revenues going forward.

We also get the following distribution among segments.

So basically EGM will continue gaining share versus media, though the largest grower will be other services (AWS, mostly). All of these assumptions and expectations are broadly in light with the optimistic projections made for With these assumptions, this is what happens to's CSOI margins if the assumptions, which described's historical margins so well continue to hold:

(click to enlarge)

As we see, what this means is that using unchanged assumptions regarding the profitability of's different segments - assumptions which have been in place even before started having profitability problems -'s CSOI profitability will continue to be low for the foreseeable future.

Absolute CSOI dollars will increase to around double by 2020, but the overwhelming part of these CSOI dollars will be captured by employees. After all, this is how stock-based compensation as a % of CSOI looks like, both until now and in the future (the assumption is that SBC will be at 1.35% of revenues in the future - which is lower than recent observations).

(click to enlarge)


I believe this model, by fully explaining the recent moves in's profitability, reflects the underlying profitability of each of's segments and of as a whole. The changing mix between these segments fully tracks the changes in profitability experienced, both in its best years up to 2010, and ever since then. The only exceptions were the AWS launch, which imposed temporary technology costs, and the Kindle eReader, which for a while sold at untypical (for large margins. Both were temporary and as soon as their effects were gone's profitability quickly fit the model again.

Using a set of assumptions regarding's future revenue growth, we can thus grossly estimate how the underlying profitability for the whole of is going to change. And what we observe is that far from displaying the much higher profitability expected by all analysts, what will deliver will be similar to what it delivered in the last 2-3 years. That is, low profitability, even when measured through's optimistic CSOI.

We thus can fully expect's profitability to continue its slow downward trajectory from here. And what's more, we can also expect employees to continue absorbing the overwhelming majority of whatever profitability is left, through stock-based compensation (SBC).

The model sees a doubling of CSOI up to 2020, but net profits should be stagnant until then due to SBC. Net profits per share should remain at or below $1.10 per share until 2020 even with massive revenue growth.

On the other hand, believing that there's a pot of earnings gold at the end of the rainbow requires the historic relationships inferred here to stop working. This seems much more unlikely given that these relationships held through moments of high and low profitability for The more likely explanation is simply that the mix matters and is still moving unfavorably, so will not see much improved profitability in the future (though the model does support up to around $1.00 in yearly earnings under the present mix, which is higher than observed recently).

In a short phrase,'s lack of profitability is structural and deeply connected to its changing sales mix. This low profitability will not change with hugely increased revenues.

Disclosure: I have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Additional disclosure: I have an options position, which would stand to gain if AMZN fell.

Source: The Lack Of Profitability Is Structural