Tag Archives: learning curve

February 28, 2015 · 8:00 am

Boeing 787 recurring costs vs. recurring income

Few days ago I was discussing with some commenters on the blog of aviation analyst Scott Hamilton (Leeham News and Comment) about the recurring costs Boeing may be experiencing in the 787 program at the moment.

I used in the discussion the analysis I had made of the learning curve Boeing has experienced in the last 2 years according to cost reductions reported by its CFO, Greg Smith. See a post I wrote about it here. The result I reached is that lately they achieved a 87%. With information disclosed last year, the figure I arrived for 2013 was ~84%, see the post here. However, in the calculation to obtain the learning curve experience the actual costs are not needed, it is sufficient to know cost reductions achieved (reported) between given units.

However, when in 2011 I wrote a series of posts (1) about Boeing 787 break even, I did try to estimate what the cost of the first production units were using published information at the time. At that time it was disclosed that Boeing had about 18bn$ of work in process (WIP) and a number of aircraft in different stages of production. Reported average costs ranged from 250m$ to 400m$. I made some simple assumptions and arrived at an average cost of 310m$ for the first ~60 units.

The next step is to accommodate those average costs into a learning curve profile. The steeper the curve (75%) the more expensive the first unit would have been. Since in 2013 the calculated curve was a 84%, I obtain that the first must have been around 650m$ (2). From then on, I apply the mentioned 84% through end 2013. Then I switch to a 87% curve (slower learning) following the reported figures from Greg Smith.

This discussion so far gives an idea of how to estimate the recurring costs. At the end of 2014 this figure is estimated around 180m$.

In order to know by when Boeing will turn the production of 787s into something profitable, we first need to know by when the recurring costs will be lower than the recurring income. The latter is estimated from the information about prices (published by Boeing here) and discounts applied (estimated in other blog posts, see the last update for 2014).

Boeing list price for the 787-8 in 2014 was 218.3M$.

These list prices are, however, increased almost on a yearly basis by Boeing. Sometimes very steeply (+11.4% in 2010, from 2008) and other times more moderately (+2.4% in 2013 vs. 2012). Going into the future I assumed this increase to be constant and about equal to 2014′ increase, 3%.

On the other hand, Boeing applies some discounts to its customers. These are never disclosed. Some are reported by some sources. What I do is to try to estimate an average discount from reported information. See a detailed calculation here. The latest figure that I arrived at was about 47%. Going into the future I assumed this discount to remain constant. You can see here the recent evolution of discounted 787-8 prices.

With all these ingredients, the only thing left is to plot together the recurring costs and recurring income:

787 recurring cost vs. recurring income evolution.

787 recurring cost vs. recurring income evolution.

As you can see recurring costs may be lower than recurring income at the end of 2019.

This will truly depend on the learning curve achieved, the number of units produced (3) and the pricing power Boeing manages to have. If the learning is steeper, the date will be sooner. If the ramp up is higher, the date will be sooner. If the discounts are lower or the list prices increased more, the date will be sooner. In any other case, either 2019 or beyond.

—

(1) See the complete series here: “Will Boeing 787 ever break-even?“, “More on Boeing 787 break even” and “787 Break Even for Dummies“.

(2) We will never know that figure. I wonder whether this is even known or registered (if not deleted and forgotten) within Boeing.

(3) For the numbers of units built I based the model in reported information that the ramp up to 12 aircraft per month is expected for 2016. I assumed that in 2015 they are at somewhere between 10 and 12 aircraft per month.

Leave a comment

Filed under Aerospace & Defence

Tagged as , , , , , , , ,

February 12, 2015 · 1:10 am

Learning Curves: Boeing 787 case (from Dec 2012 to Dec 2014)

Last year, I wrote in a couple of posts an introduction to the concept of the learning curve and provided a the case of the Boeing 787 in 2013 (based on figures disclosed by Boeing CFO, Greg Smith). In the first of those posts, I discussed that:

Boeing does not disclose outright what is the actual learning curve it is achieving in its program. Nevertheless, in its investor relations conferences it provides information here and there of cost savings achieved, etc. This can be interpreted as derived from learning curve effects, and would permit to build a model, even if based on scarce information.

This year again, let me quote again one of those excerpts from Boeing executives in order to update the exercise:

“On the 787-8, we’ve seen a decline in unit cost of approximately 30% over the last 175 deliveries and on 787-9, we’ve seen declines of 20% since the first delivery. Based on this progress, our production schedule and planned productivity investments, we continue to expect the 787 to be cash positive during 2015 and we still anticipate deferred production to decline shortly after we’ve achieved the 12 per month production rate in 2016. No change to these fundamental milestones.” Greg Smith, Boeing EVP – CFO at Q4 2014 Earnings Conference

As I mentioned last year, to the avid reader, and knowledgeable and savvy analyst, this paragraph is enough to deduce an actual learning curve achieved by the 787 for each aircraft model, provided that information reported by Boeing CFO, Greg Smith, was accurate.

Through the end of 2014, Boeing had delivered 212 787-8 and 10 787-9. The 787-9 were all delivered in 2014. The 212 787-8 were delivered as follows:

  • 2011: 3 aircraft.
  • 2012: 44 a/c.
  • 2013: 63 a/c.
  • 2014: 102 a/c.

Greg Smith is mentioning a cost reduction of 30% over the last 175 deliveries. As he was talking at the 2014 earnings call, we can safely assume that the figures were put up for him using deliveries and costs only through the end 2014 (and not counting info from January 2015 deliveries). That means that the cost improvements are measured theoretically from the aircraft 37th through the 212th (175). You can extract a report from Boeing website of the complete 787 deliveries here. I did it. Analysing it you see that between the beginning of December 2012 and the end of December 2014 176 787-8 were delivered (1).

Thus, the exercise to find out what learning curve Boeing achieved during that time span is as simple as to see what learning curve yields an “approximately 30% unit cost improvement” from the unit 36th (2) to the unit 212th. The beauty is that we do not even need to know the initial unit cost to perform the calculation, as the relative improvements in terms of percentages are independent of the starting point. All the information has indeed been provided by Boeing.

As I did in the blog post of the last year, in this first graph below I just plotted some generic learning curves, from 95% to 75%. This form of representation provides a good view of how learning is intense at the beginning of the production process and it stabilizes later on. It also shows well how learning is more intense and cost reductions are bigger for a 75% curve than for a 95% curve.

Generic learning curves.

Generic learning curves.

However, in the previous curves it is difficult to distinguish the 36th and the 212th units which are needed for the calculation. Thus, I plotted the same curves in a log scale for the numbers of units produced in the graphic below:

Boeing 787 learning curve over 176 units through Dec 2014 calculation, delta unit cost between 36th & 212th units.

Boeing 787 learning curve over 176 units through Dec 2014 calculation, delta unit cost between 36th & 212th units.

In this second graphic I added the information of what relative cost reduction is achieved between the 36th and the 212th units for each of the curves (3).

To make sure that readers are not lost, let’s take the 87% curve (in red). Following that curve, the unit cost of the 36th unit produced is a 48.7% of the 1st unit cost, whereas the unit cost for the 212th unit produced is a 34.1% of the 1st unit cost. The difference is then 48.7% – 34.1% = 14.6%, which represents a 30.0% cost reduction from the 36th unit cost. If you follow the same calculation for each of the curves, you will obtain the following unit cost improvements between 36th and 212th units:

  • 95% curve: -12.3% unit cost improvement
  • 90% curve: -23.6% unit cost improvement
  • 87% curve: -30.0% unit cost improvement
  • 85% curve: -34.0% unit cost improvement
  • 80% curve: -43.5% unit cost improvement
  • 75% curve: -52.1% unit cost improvement

Thus, from the information provided by Boeing of units delivered and unit cost improvement (“approximately 30%″, Greg Smith) we can deduce that from December 2012 through December 2014 the average learning curve that the 787 program has achieved is about 87%. Thus, in line with aerospace average indicated by NASA (85%), or in line with the reported 84% achieved in the 777. Though a bit lower than that calculated for 2013 (see here last year’s post). We can interpret that mismatch as either as a minimum error inherent of these estimates (85% vs. 87%) or that indeed the rate of improvement has somehow slowed down in 2014 in comparison to 2013.

Had Boeing been able to achieve effectively a 85% curve in the last 175 deliveries, bear no doubt that the message from Greg Smith would have been along the lines of an approximate 35% cost reduction (being the mathematical result 34%), instead of the reported 30%.

All these numbers refer to the 787-8. Now, remember that Boeing CFO also indicated that “on 787-9, we’ve seen declines of 20% since the first delivery”. 

Here, the calculation that needs to be done is the same. 787-9 deliveries started in 2014. If we assume that Greg’s comment referred only to 2014 deliveries (not 2015 ones), then the calculation must be made on 10 787-9 delivered between June and December 2014.  Here the learning curve obtained is a 93.5% ~87-88% (5) (6).

—

(1) I guess Greg Smith rounded to 175, but his “decline in unit cost of approximately 30% over the last 175 deliveries” refers to actually some fixed internal monthly reports and is based on 176 deliveries (not 175). (4)

(2) Taking into account the note (1), I will use 176 aircraft and deliveries from the 36th aircraft through the 212th.

(3) Bear in mind what a relative cost reduction is in contrast to the fact of relative costs represented in the vertical axis as percentages of the initial cost (100%).

(4) The taking of 175 instead of 176 aircraft practically does not change the result. For the same 87% curve the calculated cost reduction in one case is 30% and in the other 29.6%.

(5) This is in line with the slow down of the learning curve for the 787-8 comparing 2013 result with the result along the time span December 2012-December 2014.

(6) I initially calculated 93.5% as a result of a 20% cost reduction between the 1st and 10th units produced. However, as Matt B pointed, the first 3 787-9 have not been delivered (though they are units produced from which there was a learning effect). When you calculate what is the result for a 20% cost reduction between units 4th and 13th (the first 10 delivered), you get a curve between 87-88%, more or less the same one than for 787-8. Thus, it seems than at the beginning of the 787-8 production they went along a 85% curve that has slowed to a 87% for both models.

7 Comments

Filed under Aerospace & Defence

Tagged as , , , , , , , ,

March 5, 2014 · 8:00 am

Learning Curves: Boeing 787 case in 2013

In the previous two posts I introduced the concept of learning curve and provided a case in point (based on figures attributed to A350 FAL by Leeham News). In the first of those posts, I discussed that:

Boeing does not disclose outright what is the actual learning curve it is achieving in its program. Nevertheless, in its investor relations conferences it provides information here and there of cost savings achieved, etc. This can be interpreted as derived from learning curve effects, and would permit to build a model, even if based on scarce information.

Let me quote again one of those excerpts from Boeing executives:

“We continue to see progress in key operational performance indicators and unit costs, as we further implement production efficiencies and stabilize the overall production system on the 787 program. Unit cost has improved approximately 20% over the past year on the 787-8 […]“, Greg Smith, Boeing EVP – CFO at Q4 2013 Earnings Conference.

To the avid reader, and knowledgeable and savvy analyst, this paragraph is enough to deduce the actual learning curve achieved by the 787 during 2013, provided that information reported by Boeing CFO, Greg Smith, was accurate.

During 2013 Boeing delivered 65  787s, from the 50th to the 114th units (in previous years it had delivered already 3 in 2011 and 46 in 2012).

Thus, the exercise to find out what learning curve Boeing achieved in 2013 is as simple as to see what learning curve yields an “approximately 20% unit cost improvement” from the unit 50th to the unit 114th. The beauty is that we do not even need to know the initial unit cost to perform the calculation, as the relative improvements in terms of percentages are independent of the starting point. All the information has indeed been provided by Boeing.

In the first graph below I just plotted some generic learning curves, from 95% to 75%. This form of representation provides a good view of how learning is intense at the beginning of the production process and it stabilizes later on. It also shows well how learning is more intense and cost reductions are bigger for a 75% curve than for a 95% curve.

Generic learning curves.

Generic learning curves.

However, in the previous curve it is difficult to distinguish the 50th and the 114th units which are needed for the calculation. Thus, I plotted the same curves in with a log scale for the numbers of units produced in the graphic below:

Boeing 787 learning curve in 2013 calculation, delta unit cost between 50th & 114th units.

Boeing 787 learning curve in 2013 calculation, delta unit cost between 50th & 114th units.

In this second graphic I added the information of what relative cost reduction is achieved between the 50th and the 114th units for each of the curves (1).

To make sure that readers are not lost, let’s take the 85% curve. Following that curve, the unit cost of the 50th unit produced is a 40.2% of the 1st unit cost, whereas the unit cost for the 114th unit produced is a 32.9% of the 1st unit cost. The difference is then 40.2% – 32.9% = 7.2%, which represents a 18.0% cost reduction from the 50th unit cost. If you follow the same calculation for each of the curves, you will obtain the following unit cost improvements between 50th and 114th units:

  • 95% curve: -6.1% unit cost improvement
  • 90% curve: -12.0% unit cost improvement
  • 85% curve: -18.0% unit cost improvement
  • 80% curve: -23.8% unit cost improvement
  • 75% curve: -29.6% unit cost improvement

Thus, from the information provided by Boeing of units delivered and unit cost improvement (“approximately 20”, Greg Smith) we can deduce that during 2013 the learning curve that the 787 program has achieved is between 85% and 80%. Thus, in line with aerospace average indicated by NASA (85%), or in line with the reported 84% achieved in the 777.

If we wanted to know what learning curve yields exactly that 20% unit cost improvement, it is now trivial to calculate it: the 83.3% learning curve.

Having made these numbers, and taking into account the words used by Boeing CFO, “unit cost has improved approximately 20% over the past year on the 787-8″, I take it as that the improvement has been close to 20% though probably not reaching it; thus, I understand that the learning curve was rather between 83.3-85% instead of down to 83.3%.

—

(1) Bear in mind what a relative cost reduction is in contrast to the fact of relative costs represented in the vertical axis as percentages of the initial cost (100%).

2 Comments

Filed under Aerospace & Defence

Tagged as , , , , , , , ,

March 4, 2014 · 8:00 am

Learning Curves (case in point)

In a previous post I made an introduction of the concept of learning curve and the gains that can be achieved through experience, with the focus on aerospace industry.

Case in point 1

Few days ago I found an interesting blog post at Leeham News, “Lessons learned from A380, 787 benefit A350“, which discusses lessons learned reportedly being applied into the A350 programme and includes two interesting tables. I will use one of those tables in order to show an exercise of learning curve calculation, without caring of whether the information included in the table is accurate or reflects actual lead times of A350 at Final Assembly Line (FAL). [I do not validate the information of the table despite of using it for an intellectual exercise.]

A350 days in FAL as reported by Leeham News (Reprinted with permission by Leeham Co).

If we take the lead times reported by Leeham News in the table above, we can calculate to what learning curve profile would those lead times correspond. See the result in the graphic below:

Learning Curves.

Learning Curves.

The information reported by Leeham News corresponds to a learning curve between 85% and 90%, about 88%. If figures resemble the reality, between the first and second units, the learning was closer to 95%, however, between the 2nd and 3rd it improved greatly. Since the 3rd unit, the learning is more stabilized at about 88%. [Again, having made this calculation as an exercise does not validate the information taken as an input]

It is important to bear in mind the units used in this calculation are days. Days of assembly are linked to unit production costs, but not necessarily in a linear fashion, as it will depend on the labor / automation being employed in the production of each unit, which may change as production evolves. It is also important to note, that the example of learning curve calculated would refer only to assembly at FAL and not to the overall production process, from material costs, production at earlier stages of the manufacturing process, etc. The value added at FAL is but a minor percentage of the value of an airplane.

I leave for a following post a case in point 2, with the learning curve of the 787 during 2013.

1 Comment

Filed under Aerospace & Defence

Tagged as , , , , , ,

March 3, 2014 · 8:00 am

Introduction to Learning Curves

Let me introduce the learning curve effect by quoting directly from the Wikipedia:

“The rule used for representing the learning curve effect states that the more times a task has been performed, the less time will be required on each subsequent iteration. This relationship was probably first quantified in 1936 at Wright-Patterson Air Force Base in the United States, where it was determined that every time total aircraft production doubled, the required labour time decreased by 10 to 15 percent.” […]

“Learning curve theory states that as the quantity of items produced doubles, costs decrease at a predictable rate.”

I used the concept of learning curve in a blog post in which I discussed whether and when the Boeing 787 would break even. In that post I referred to Boeing’s target of reaching a 75% learning curve on the 787 program, much more difficult to reach than the 84% that Boeing reportedly achieved in the 777 program.

Why is a curve of 75% more difficult to achieve than a 84% curve? The meaning of the figure “84%” attached to the learning curve is that each time that the number of units produced is doubled, the cost is reduced in 16%, or the 2*nth unit cost is 84% of the unit cost of the nth unit. Thus, a 75% curve would imply that the cost is reduced in 25%, which is a higher cost reduction than 16%, and, thus, more difficult to achieve.

On the other hand, NASA, in its Learning Curve Calculator, offers some guidance on learning curves for different industries and mixes of hand labor and machining work:

  1. Aerospace 85%
  2. Shipbuilding 80-85%
  3. Complex machine tools for new models 75-85%
  4. Repetitive electronics manufacturing 90-95%
  5. Repetitive machining or punch-press operations 90-95%
  6. repetitive electrical operations 75-85%
  7. Repetitive welding operations 90%
  8. Raw materials 93-96%
  9. Purchased Parts 85-88%

How is the concept of learning curve calculated? (from Wikipedia: )

 Now the equation for the unit curve is given by:

Y_x = K x^{\log_2 (b)}

where

  • K is the number of direct labour hours to produce the first unit
  • Yx is the number of direct labour hours to produce the xth unit
  • x is the unit number
  • b is the learning percentage (expressed as a decimal)

How does a learning curve look like? (from Wikipedia) Actual examples of curves in both linear and logarithmic scales would be:

Experience curve (from Wikimedia, by Apdevries).

The concept of learning curve is indeed used in aerospace, however, coming back to the 787 program, Boeing does not disclose outright what is the actual learning curve it is achieving in its program. Nevertheless, in its investor relations conferences it provides information here and there of cost savings achieved, etc. This can be interpreted as derived from learning curve effects, and would permit to build a model, even if based on scarce information.

See some of the hints that Boeing provides:

“We continue to see progress in key operational performance indicators and unit costs, as we further implement production efficiencies and stabilize the overall production system on the 787 program. Unit cost has improved approximately 20% over the past year on the 787-8 […]”, Greg Smith, Boeing EVP – CFO at Q4 2013 Earnings Conference.

“[…] as we continue to make improvements 787 unit cost […]”, Greg Smith, Boeing EVP – CFO at Q4 2013 Earnings Conference.

“when you look at flow-time, you look at unit cost at Charleston whether it’s final, mid, or (aft) it made great progress there. And the team has been very focused on continuing that progress going forward. We have experienced a higher number of jobs behind schedule in the mid-body section, and that’s really due to, if you think about it, you are introducing the Dash-9 at the same time going to 10 a month. […], we’ve applied additional resources. We know how to do this and we’ll get those jobs back to what we view as a more acceptable level. So we got mitigation plans.” Greg Smith, Boeing EVP – CFO at Q4 2013 Earnings Conference.

“This morning we announced plans to increase 787 production beyond the 10 per month we’re on track to achieve this year to 12 per month in 2016 and then 14 per month, before the end of the decade. […], capture productivity and learning improved profitability […]” Jim McNerney, Boeing Chairman, President and CEO at Q3 2013 Earnings Conference.

“We’ve added another line or sorry, a position within the line, where we’re doing the wing, body joint earlier in the process and this is through experience after 134 airplanes, the teams are really coming up with better ideas or improvements on how to increase flow and that’s going to require some upfront investment. But obviously in the units to come after we’ll see that improvement again in flow and productivity.” Greg Smith, Boeing EVP – CFO at Q3 2013 Earnings Conference.

“[…] the flow time reductions, we’ve had in our factories, the hours per unit, the productivity per whatever are increasing significantly on all of our programs.” Jim McNerney, Boeing Chairman, President and CEO at Q3 2013 Earnings Conference.

2 Comments

Filed under Aerospace & Defence

Tagged as , , , , , , ,

October 31, 2011 · 2:00 pm

More on Boeing 787 break even

After the post I wrote the last week with an analysis of when could Boeing 787 break even (Will Boeing 787 ever break even?), I received some feedback in the form of emails, comments in the blog and comments in Scott Hamilton’s blog (Leehamnews) as he mentioned the analysis and linked to it.

I wanted to address some of those comments in a single piece for the benefit of all. Please, find them below.

A leading aviation analyst hinted that:

“In the case of the 787, it increasingly looks like the -8 will be an up-front version with poor pricing, but the -9 (and -10) are planned to make this problem better. […] So, your assumption of a consistent 38% discount  won’t happen, unless either Boeing fails to improve the 787’s performance or if the market doesn’t like the 787.”

This could be built into the model by allowing aircraft sold after the current backlog appear in the model as sold at a lower discount (e.g. 20% instead of 38%). The cash inflows due to the down payments would be increased soon, but since producing the existing backlog will take until sometime in 2019, the cash inflows from deliveries will be untouched until then. Result: that would bring break even about 2 years forward (2021 vs. 2023, in the case of 75% learning curve and 10% discount rate – what I would call “Boeing’s baseline”).

Another leading analyst suggested:

“I think the 5,000 market forecast is for the middle twin-aisle market as a whole, which includes A330, A350 and 777 (and even the almost-but-not-quite-dead 767). If I’m correct, I disagree with your figure above as too high. Clarification of the 5,000 is required for everybody.”

I reviewed Boeing’s commercial market outlook. In the next 20 years (2011-2030) Boeing sees deliveries of 6,610 twin-aisle passenger aircraft: 3,020 small and 3,590 intermediate.

In the previous post I mentioned that 2,634 were included as 787 deliveries, though that includes deliveries from 2031 to 2034, later than the period covered by this year’s CMO (480 a/c between 2031-34).

To compare apples with apples, between 2011 and 2030 I included 2,154 787s delivered, all those that 787 assembly lines could produce. That is 43% of the “addressable” 5,000 market or 33% of the 6,610 total market, including A330s, 777s (70-90 produced a year now), 767s (12-15 a year) and A350s… This figure, 2,154, could be optimistic in my opinion as well. If a lower figure should be used the situation for the break even would be worse.

I received some comments via Leehamnews blog:

The user KDX125 mentioned:

“[…] the current inventory of 18bn is distributed over 58 aircraft that are assumed to be WIP. But the amount should be limited to deferred production cost and unamortized tooling, which according to the 10-Q is ‘just’ ~11bn.”

The 10Q says (emphasis is mine):

“As of September 30, 2011 and December 31, 2010, commercial aircraft programs inventory included the following amounts related to the 787 program: $14,423 and $9,461 of work in process (including deferred production costs), $1,775 and $1,956 of supplier advances, and $1,770 and $1,447 of unamortized tooling and other non-recurring costs. As of September 30, 2011, included in work in process were deferred production costs related to the 787 program totaling $9,699.”

Those are the ~18bn$ I mentioned, which were mentioned in the conference call as well. We need to distinguish here between accounting and cash flows. For the accounting of the profit behind each aircraft, it may be true that unamortized tooling could be distributed among 1,100 aircraft, however all those are costs that have already meant an outflow of cash. What I tried to do is to see how much costs of the about 50 aircraft which are in different stages of production was included in those 18n$ in order not to double count cash outflows related to costs of aircraft delivered. In other words, if from those 18bn$, 1.8bn$ refer to tooling and shouldn’t be distributed as WIP of those ~50 aircraft, that means the cash flow profile would look even worse, not better.

Normand Hamel and others have mentioned:

“[…] not taken into account the penalties to the customers and various suppliers.”

That’s correct and I acknowledge this is a shortcoming of the model. I tried to base all assumptions in public references appearing either in Boeing’s website or reports or news in the media. I couldn’t find anything related to value and structure of those penalties and thus didn’t include them. I would welcome references regarding this point. Anyway, as it was mentioned, including penalties in the model would worsen the cash flow profile and delay break even.

Slinger raised the point:

“For the learning curves to work in this model one must now the relationship between labour cost and material cost, since the learing curve only applies to labour time. […]

I don’t know if this is factored in into the model but if not, it would make the forecast even more pessimistic.”

While it is true that generally the concept of the learning curve is applied to labour, C. Lanier Benkard (professor of Economics at Standford Graduate School of Business), in his paper “Learning and Forgetting: The Dynamics of Aircraft Production” [PDF], describes the learning process as follows:

“Learning may take on many different forms depending on the particular nature of production. In more capital-intensive industries such as chemical processing and semiconductors, learning primarily results from the fine-tuning of production techniques. In such industries, engineers and managers analyze current output and constantly make small changes to the process, with the result that productivity gradually improves. In labor-intensive industries such as aircraft and shipbuilding, learning primarily results from workers becoming more efficient at the tasks the perform through multiple repetition. Many industries may be subject to both types of learning. […]”

Since I haven’t found any source referring to different types of learning for labour and materials for the case of Boeing, I made the same simplification that professor Benkard does in his paper (emphasis is mine):

“If the Leontieff assumption in equation (2) were relaxed and instead production was assumed to be Cobb-Douglas in all inputs, the production only at the unit level would imply a labor-requirements equation similar to (4) with the addition of both the wage rate and materials prices […]”.

If this assumption was optimistic or if the learning related to the materials production was capitalized mainly by Boeing’s suppliers and not by Boeing itself, this would only make the cash flow profile worse and delay the break even date.

Finally, Garry Reinhardt in the comments section to my post asked:

“[…] is Boeing profit, going forward, being decreased or increased by each 787 delivered? And if it’s negative now, when will it reach zero (leaving out the previously paid for expenses)?”

I do not give stock recommendations. Anyway, this cash flow analysis shows that most of the cash outflows that make the case of the 787 (viewed in isolation) such difficult, were made in the past years and didn’t derailed Boeing then. The 787 development was supported by other Boeing programs. From now on what is missing is that unit production cash outflows are lower than cash inflows related to deliveries (penalties not taken into account), and that is something that should happen at sometime between 2014 and 2016, depending on the learning curve that Boeing achieves. Nevertheless, whether this is something already reflected or not in current Boeing price share, and whether this price is cheap enough or not is left to the individual investor.

Finally, thanks again for all your feedback, comments, criticism and especially to Scott Hamilton for his linking the post so that I could get more feedback.

12 Comments

Filed under Aerospace & Defence

Tagged as , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

October 28, 2011 · 12:41 am

Will Boeing 787 ever break-even?

Boeing 787, the Dreamliner, made its first commercial flight with Japanese airline All Nippon Airways between Tokyo and Hong Kong. The event has been widely covered in general and specialized press. One of such articles, in Bloomberg, was the spark of a conversation with a colleague that led to the following question: will the 787 ever bring value to Boeing? Will the net present values of all the cash flows related to the programme ever be positive?

After a conversation, I decided to try to answer to those questions, using information found in different sources across the internet.

Lately there has been much discussion in the media about what would be the Boeing’s “accounting block” size (finally it was announced to be 1,100 aircraft units). This concept refers to the number of aircraft upon which Boeing is going to spread the amortization of the work in process now in the balance sheet (around 18bn$ at 2011 Q3, according to Boeing), excluding R&D costs (which were already accounted for in previous years’ income statements).

While the concept has raised much attention, it is irrelevant to appraise the 787 as a long-term investment project, for which yearly cash flows, accordingly discounted for (to take into account the time value of money), shall be calculated. This time including R&D costs.

Together with the previous concept, last days’ news have often discussed the concept of “learning curve”. This is a concept typically used in aerospace industry and which will be central to the discussion of whether the 787 will be or not a success as an investment project.

Learning or experience curve

Let me introduce the learning curve effect by quoting directly from the Wikipedia:

“The rule used for representing the learning curve effect states that the more times a task has been performed, the less time will be required on each subsequent iteration. This relationship was probably first quantified in 1936 at Wright-Patterson Air Force Base in the United States, where it was determined that every time total aircraft production doubled, the required labour time decreased by 10 to 15 percent.” […]

“Learning curve theory states that as the quantity of items produced doubles, costs decrease at a predictable rate.”

The press has stated that Boeing’s targeted learning curve for this programme will be 75%, despite of having less control over the supply chain in comparison with previous developments. That means that every time production units have doubled, the unit cost will have decreased in 25% (if unit 100th costs 100, unit 200th shall cost 75).

In the last new aircraft development, the 777, Boeing reportedly experienced a curve of 84% (costs decreasing 16% every time production units doubled). In this exercise I will make use of different values for the learning curve in order to see its influence (90%, 84% -as with 777-, 80% and 75% – Boeing’s target).

Discounted cash flows

Due to the time value of money, i.e. the interest that could be earned by a given amount of money, it is important to evaluate the present value of the cash flows of the 787 project along its life-cycle. As the Wikipedia states it:

“Net Present Value is a central tool in discounted cash flow (DCF) analysis, and is a standard method for using the time value of money to appraise long-term projects.”

This basic concept of finance theory is rarely covered by the press.

The then-assistant professor of Economics and Public Affairs at Princeton University, U.E. Reinhardt, in his paper “Break-even analysis for Lockheed’s TriStar: an application of financial theory” (PDF, 2001), analyzed comprehensively the project of the TriStar making use of information that became public at a moment when Lockheed officials took part in Congressional hearings over a loan-guarantee needed by the TriStar programme (1971).

Reinhardt found that the figures that appeared in the media led one to consider that the programme cash flows were not discounted either to the time at the beginning of the project  or to the time of the congressional hearings, as prices and market numbers would had only produced a positive net present value if the discount rate had been in fact around 0%, that is “only if one assumes that the company was prepared to advance the enormous sums required for that project without asking for any positive return on this investment”.

Typically the discount rate used to evaluate different projects is the cost of capital of the company. For this exercise I have used different rates to see its influence in the results (0%, 5%, 10% and 12%, being the last two typical figures used in industry).

As a side note: Reinhard used as learning coefficient 77.4%, close to the optimistic 75% targeted by Boeing and lower than the disclosed 84% of the 777 case.

Data gathering

In order to build the exercise, different sets of data need to be collected. I will discuss them below, indicating the sources used and explaining the assumptions taken.

Number of aircraft produced

According to news reports, this year Boeing will deliver between 15 – 20 787 and 747-8, being about two-thirds of the latter. From that information I took that about 6 787s will be delivered in 2011. From then on, the ramp I used tries to replicate what the media is reporting.

Boeing intends to reach a rate of 10 aircraft per month by end 2013, thus from 2014 I assumed Boeing will produce 120 787s every year (as Boeing factors down time into calculation of the monthly rate).

According to Seattle Times, this week Boeing is increasing its rate from 2 aircraft per month to 2.5. Since between 2012 and 2013 this rate has to increase from 2.5 to 10, I assumed an average of 3 aircraft/month for 2012 and 6 for 2013.

Number of aircraft sold

Boeing publishes in its website the number of aircraft ordered each year. For this exercise, I took into account the net orders in the year of order.

In the last conference call, Boeing stated that it sees an addressable market for the 787 in the next 20 years of 5,000 aircraft (this number of aircraft reflects deliveries). As of today, Boeing has close to 800 orders. This backlog covers the production until somewhere in 2019, in order to keep the production line with a steady production rate, I assumed Boeing will sell the necessary aircraft to allow a steady state production of 120 aircraft per year until the end of the exercise in 2034. That would mean Boeing would have delivered 2,634 aircraft, a bit over half of the 5,000 aircraft that represents the addressable market.

Since Boeing has already sold close to 800 of those 2,634 aircraft, I assumed the rest will be sold evenly every year, or at a rate of 82 aircraft per year from 2012.

List price of the 787

Boeing publishes in its website the list prices of all its models. The list price of the 787 today ranges from 193.5 to 227.8 million USD, or about 211 M$ on average.

At the end of 2010, Boeing raised its list prices 5.2% on average. The previous raise had taken place in 2008, 2.6% in relation to 2007 prices, when they had been raised another 5.6% after two years.

From those numbers, I made the assumption forward and backward that on average Boeing raises its list prices about 2.6% per year.

Price discounts

I have already published two different posts about average Boeing price discounts for 2009 and 2010. A recent article from Flight Global confirmed the order of magnitude for the case of the A380. The discount rate I used for the exercise is 38%.

Another confirmation of the order of the discounts specifically for the 787 comes from the following analysis by Jon Ostrower, from Flight Global, who reports prices of 787 to be around 76M$ in 2004-2006, excluding engines, which were about 20-30M$ (a total of 96-106M$).

Taking the list prices above for those years and using the 38% discount for that period of 2004-2006 the real prices given by my model are in the order of 101-104M$, in the same order that Jon’s information.

Down payment

Here I used for the exercise the same assumption I had used to calculate Boeing discounts in previous posts: a single 3% down payment taken from the AIAA paper “A Hierarchical Aircraft Life Cycle Cost Analysis Model” by William J. Marx et al.).

Regarding down payments, I once received another input in my blog from the analyst Scott Hamilton (Leeham), where he mentioned several progress payments of 3-5% of the price of the aircraft so that at the time of delivery 30% of the aircraft had already been paid for. Simulating these payments complicates the model, but, since early cash inflows may have an impact in the break even analysis, I have checked the variation with different sizes of down payment (3%, 20%).

Costs

These are the main inputs needed for the whole analysis. Most of the non-recurring cost (NRC) have already been incurred, thus, there is little room for manoeuvre in there. However, regarding the recurring costs (RC), those are where Boeing has the chance to make the programme profitable.

NRC: Research & Development (R&D) and capital expenditures (CAPEX)

Seattle Times provided a detailed description of the incurred costs that Boeing had through end September 2011. The account was split in different categories of non-recurring costs (R&D, CAPEX, buying out partners) totalling over 16bn$.

RC: Inventory, advances to suppliers

In the same article there was another explanation about the recurring costs (inventory –work in process, supplier advances and others) incurred so far: up to 16.3bn$ by September. This last figure has risen to 18bn$ according to the last conference call given by Boeing.

In order to build the cash flow profile, it is necessary to know the cash flow profile of the costs described above. Since we have only information regarding some of those (buyout of partners, CAPEX for Charleston FAL) I needed to make an assumption for the rest.

In this case, I used the same cash profile used by B. Esty and P. Ghemawat (both then at Harvard Business School) in their paper “Airbus vs. Boeing in Super Jumbos: A case of failed Preemption” [PDF], where they performed a valuation analysis for then known as A3XX (now A380).

You may see in the following graphic the cash profile used up to now:

Boeing 787 costs through 2011 profile.

Regarding the recurring costs, the main finding to be done is the recurring cost of the aircraft at a certain point. Once we have a reference, we can play with the learning curve to see how the costs will be in the future or were in the past.

Different sources quote figures as 250-300M$ or even 400M$, but they do not explain whether that is the cost of a unit right now (e.g. serial number 44) or whether it is the average cost up to now (they do not indicate how many aircraft are included in that average either if that was the case).

The approach I followed was different.

Boeing has disclosed it has 18bn$ as WIP related to the 787, and from local press we know what is the state of production right now, thus, we can try to estimate what is the average cost of the aircraft already produced.

The aircraft delivered to ANA needs to be discarded as its portion of WIP has already been included in the income statement of Q3. We know that at Seattle FAL there are now serial numbers 46 (to be completed according to press) to 50 (just arrived, according to Jon Ostrower). Thus, within WIP there must be about 44 aircraft almost completed and another 5 about 98% completed (FAL value added is around 4% of RC).

As we do not know how many other aircraft are in process and at which stage each one, we need to make further assumptions. From the same article of Seattle Times we learnt the following:

“The first 40 out of the Everett factory required massive and repeated rework, and the next 10 to 20 also need modifications because of design changes after flight testing.”

That means that at least about another 20 aircraft are being manufactured when other aircraft are at FAL. We can assume an average stage of completion of 50% of costs incurred in each one.  With these assumptions, we gather that the 18bn$ correspond to about 58 aircraft in different stages of work in process (WIP).

This gives us an average cost of about 310M$ apiece, close to the figures mentioned by some sources. The difference is that now we have a reference of cost and aircraft unit or units for that cost (1 delivered + 44 finished + 5 at FAL + 10-20 WIP ~ 60-70 aircraft).

With this average cost and using the learning curve formulas, we can deduct the unit cost of the first unit produced, which will be different for each learning curve we select, and from which all the costs of future units will be calculated.

Side note: Both 80% and 75% curves yield lower unit costs than the aircraft price (with the 38% discount) by the year 2015, in line with what was disclosed by Boeing CFO James Bell during the conference call:

“Bell projected that the cost to build each Dreamliner will drop below the price paid by the buyer around 2015, providing positive cash flow for the first time.”

Analysis

With all the data gathered above, and the required assumptions made, we can build a comprehensive valuation analysis.

Boeing 787 will not break even before 2034

The first graphic that is shown is what I believe will be the real scenario for the 787: it will not break even in the first 30 years of the programme, discounting cash flows. It won’t break even before 2034. In this case, I used the learning curve of 84%, that which was reported as the real one for the 777, and which is more conservative than the targeted one, 75%. As discount rate for the cash flows I used 10%, which could be even considered a bit low.

787 cash profile for a learning curve of 84 and discount rate of 10%.

When will then the 787 break even?

I continued the series for this set of reference parameters (84% curve and 10% discount rate) and break even would indeed happen, but not before 3650 units of the 787 have been delivered at the year ~2046. By then production rate would have had to slow down to about 6-7 aircraft per month as the backlog would have been already consumed, thus a new cost structure per unit produced could even make the mentioned date to be deferred even later.

Influence of the discount rate

The discount factor could be assimilated with the cost of capital. The reference I used was 10%, but let see how the previous graphic would look like in the case the factor was lower 5% and a bit higher 12%.

787 cash profile for a learning curve of 84% and discount rate as parameter.

We can see that if the cost of capital was as low as 5%, the 787 would reach break even by around 2028, but in the case of 12% it would never break even within 30 years as well.

Influence of the learning curve

The learning curve I used was the one I believe is more realistic as previous Boeing’s experience has shown, 84%. But let’s see how it influences the valuation, now fixing the discount rate at 10% and using the learning curve as a parameter:

787 cash profile for a discount rate of 10% and learning curve as parameter.

We can see that if the curve is the 90% the outlook is much darker, however, for 80% the programme would break even within the first 30 years, at around 2029. If the curve is the one Boeing targets, 75%, the programme may break even at around 2023, in line with Boeing statement:

“The positive cash flow will gradually pay back the earlier production costs to finally break even on manufacturing the planes roughly 10 years from now, Boeing said.”

Influence of the discount in prices

I have used a 38% discount over prices. I feel quite confident that Boeing discounts are around that figure, nevertheless, let’s see what would happen to the reference case (84% curve) if discounts were just ~20%, about half of those used, which would make cash inflows much higher at the moment when deliveries start.

787 cash profile for a learning curve of 84%, discount rate as parameter and discount over price of 20%.

We can see that in that case, the break even would be within the first 30 years, at about 2024 already for a DCF discount rate of 10%. Nevertheless, I doubt that pricing power of Boeing will allow it to stop giving ~38% discounts.

Influence of the down payments

In the simplified case that the down payment at the time of ordering the aircraft wasn’t in the order of 3% but 20%, this would bring forward cash inflows especially related to the first 800 aircraft ordered and thus improve the business case. However, once the programme is in steady state it wouldn’t change much.

You may see that maximum cumulative negative cash flow for the 5% curve only reaches about ~19bn$ vs. ~32bn$ in the reference case. Also for the 5% curve the break even is brought forward 2 years (from 2028 to 2026).

787 cash profile for a learning curve of 84%, discount rate as parameter and down payment of 20%.

“Boeing’s view” on the matter?

As we have discussed above, Boeing is targeting a learning curve of 75%, ambitious if compared to that of the 777, but not far from that used by Reinhard in his paper.

"Boeing's view": 787 cash profile for a learning curve of 75%, discount rate as parameter.

In this case, you may see that with a discount rate of 10% break even is reached in 2023. Again, as mentioned above, this is in line with Boeing’s comments in the conference call. Even for a discount rate of 12% break even would come within the next decade at around 2026.

This stresses the importance of the learning curve effect and cutting costs during the series production phase. Being in the state that the programme is now, with about -22bn$ cumulative cash flows through 2011, the only way to save it is through experience gained at the production sites.

The question then is: Will Boeing be able to achieve that 75% curve?

Side note: Let me now come back to the price discounts. I find that the results shown in the last graphic, with the curve Boeing intends to achieve (75%) and a discount rate for cash flows of 10%, as another confirmation of the discount used for prices of aircraft.

The model I built takes the 38% figure discounted from list prices, and I find it remarkable that with that figure the model predicts lower unit costs than aircraft discounted prices by 2015 (as mentioned by the CFO, James Bell) and predicts a programme break even at around 2023, about 10 years from now, as mentioned in the conference call. I take the last conference call as an implicit confirmation from Boeing of the discounts it applies to its 787s.

What has been the effect of this 3-year delay?

This is not easy to estimate, but I’ll give it a try.

Most of the assumptions remain constant. I believe that the delay has primarily deferred cash inflows from deliveries, extended R&D related to engineers working in the development for 3 years more sorting out problems and increased WIP of aircraft waiting in the production line. I will also remove cash spent in buying out partners, though this might be arguable as possibly the price of the buyout is cash neutral taking into account partners’ margins disappearing.

Below you may see the two graphics, one for the reference case (84% curve) without delay and the other for Boeing’s target (75%):

Effect of 3-year delay to 787 cash flows for learning curve of 84%.

Effect of 3-year delay to 787 cash flows for learning curve of 75%.

You may see that even in the conservative scenario (84%) break even is reached in 2014 or 2015 depending on the discount rate. With the more aggressive learning curve, 75%, the prospect is even rosier: from 2013 for both discount rates the programme would have reached break even less than 10 years since its launch in 2004 and 10 years before what now seems to be Boeing’s target break even year, 2023.

This result clearly points out how much a delayed programme may hurt the business case of an aircraft development: from being a sound project to converting it into a nightmare that may never break even, jeopardizing future developments in terms of lack of financial and human resources available.

What is the “accounting block” size used for then?

As its name points out, this is a mere accounting issue. It permits Boeing to spread already incurred costs that have been capitalized (in the balance sheet, not in previous years’ income statements) among aircraft to be delivered in the future. Since Boeing has a backlog of almost 800 aircraft and believes it will sell over 2,500 aircraft, it has prudently opted to spread costs over 1,100 aircraft, allowing itself to start reporting profits on each aircraft delivered almost from day one (from about 2015 in fact), having a shinier income statement and the bonuses that come with it.

—

Finally, as Richard Aboulafia would put it:

“Yours, ‘Til The 787 Breaks Even”

31 Comments

Filed under Aerospace & Defence

Tagged as , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,