Re: David Morrison

From: John Conover <john@email.johncon.com>
Subject: Re: David Morrison
Date: Fri, 30 Jan 1998 23:44:20 -0800


The attached has been circulated in rather wide distribution, and has
found its way into the Business Process Reengineering and Learning
Organization conference distributions, (I was prolific in these
conferences in the late 80's, and a lot of the email I am receiving is
from those folks-stuff I have left out, or forgotten.) I have been
asked to make a FAQ of the question/answer section. So, I will update
and forward stuff as it comes in. I will keep the subject the same, so
if it annoys you, that is what the 'D' button on your keyboard is for,
(the one under your left social finger.)

Q/A's 14-20 added ...

        John

David Morrison of Mercer Management Consultants has written a book
entitled "The Profit Zone." The premiss of the book is that optimizing
a company's operations for maximum market share and/or maximum
profitability are mutually exclusive propositions.  Is this true? It
turns out that it is. Here is why.

There is ample evidence that industrial markets can be modeled as
fractal time series. For theoretical arguments, see "Competing
Technologies, Increasing Returns, and Lock-In by Historical Events",
Econ Jnl, 99, pp. 106-131, 1989, which is also available on line from
http://www.santafe.edu/arthur. For empirical insight, see the
databases at http://www.stat-usa.gov/.  (You can look at the marginal
increments/returns with a spread sheet.) It is handy to have a
theoretical and empirical framework when contemplating business
concepts-and these frameworks are accurate, (and in agreement,) to
about 7 decimal places, with about 5 orders of magnitude in
scalability.

What are the implications of looking at things this way?

For starters, lets look at the chances of a company with small market
share being able to dislodge a company with large market
share. Suppose, for example, there are only two competitors in a
marketplace, and the market share is divided 99%/1%. What is the
chances that the smaller company will EVER become the market leader?
It is 1/99, or about 1 percent. And, how long, on average, (assuming
that a market share/cash inflow of zero for one month will make the
company "bust,") will the smaller company survive in the marketplace?
Its about 1 * (99 - 1) = 98 months, or, about eight years. (Kind of
says something about investing in a microprocessor company that is
going to dislodge Intel, Huh?)

And, how will the market share of many companies competing in an
industrial market arrange their selves? (Or, more correctly, what is
the frequency distribution of companies with a given market share?) It
will be proportional to 1 / N squared, where N is the the number of
companies competing in a "tier level" of an industrial market. So, we
would expect that integrating this would give fraction of market
share, which would be proportional to 1 / M, where M would be the
market share, (ie., the first company's market share would be
proportional to 1 / 1, the second to 1 / 2, the third 1 / 3, and so
on.)

This is a useful concept, since the number of companies competing in
an industrial market can measured from the market growth graph. The
number of companies is proportional to the average of the marginal
growth divided by the root mean square of the marginal growth,
(assuming the CEO's in the companies have something going for them,
and are running their companies at near optimal-ie., taking some
risks, but not too much. In point of fact, the risks should be equal
to the average of the marginal growth divided by the root mean square
of the marginal growth. Thanks for sharing that, huh?)

So, what does all this have to do with what David Morrison is talking
about?  A CEO can run a company that optimizes profitability, (where
the average of the profit margins is equal to the root mean square of
the profit margins,) or, alternative, the company can optimize market
share, (where the average of the market share increments is equal to
the root mean square of the market share increments.) But not both.

There are interesting consequences to this. Either alternative can be
stable in the long run, (but anyplace in between is not-the company
will, eventually, go bust if it attempts a compromise alternative.)
Interestingly, the semiconductor industry always optimizes for market
share-the US automotive industry always optimizes for
profitability. You can use your spreadsheet and the DOC database to
verify this interesting little tidbit, which is a handy thing to know
if you are investing in industrial companies, or going to compete in
an industrial market-which tells you what you have to do to be
successful, in either case.

Want to be a hero CEO? Well, if you run your company with the root
mean square of the marginal increments greater than their average, you
will be a hero. But only for a while. Here's why. The root mean square
of the increments is a measure of the chances taken. The average is
the returns made by taking those chances. It is not trivial to prove,
but the optimal operating point is when the root mean square, squared,
is equal to the average-and this will maximize long term
growth. Increasing the root mean square, by taking more chances, will
result in very substantial short term growth at the expense of the
long term.  (You could see the "chancy" operations of the likes of
Netscape and Ascend reflected in the value of their equities. A big
run up, followed by a "crash".)

It is not a complicated concept. Too little risk taking means too
little growth-too much risk taking means large short term growth, and
small long term growth, (another of economics most profound insights.)

But now, you know how to measure and optimize it.

        John

BTW, optimization is not a guarantee for winning. The only consolation
would be that the best possible game strategy was played. And in case
you are curious, dividing the average of the marginal increments by
the root mean square of the marginal increments, and adding one, then
dividing by two is the likelihood that an industrial market, (or P&L,)
will increase in value. As an approximation, just counting the number
of up movements in a time interval, say a hundred days, will be very
close to this number. Useful concept. (It is the information-theoretic
Shannon probability, in case you are curious, and is a measure of the
entropy of the mechanics that make industrial markets and P&L's
fluctuate.)

Want another interesting little tidbit? If we want to know how long a
company, on average, will dominate an industrial market, in years, it
would be proportional to 1 / sqrt (t), or about 4 years. (Want some
empirical evidence-look at the quarter century of the semiconductor
industry.)

More? How about how long a company will remain solvent, (again, using
a time scale of years, assuming an insolvent year means the company
would be "bust,") after doing an IPO in 5 years? Again, it is 1 / sqrt
(t), where t is 5 year intervals, or about 20 years. (Want some
empirical evidence-the average number of years that a company is on
the NYSE is 22 years.)

Or, how about the chances of a new start up succeeding in 60 months =
5 years, if a "bust" month means the company is insolvent? It is,
again, 1 / sqrt (t), where t is months, or one about 8. (Want some
empirical evidence-talk to your friendly VC. They run about one in 9.)

Some more? How about what the DOC at http://www.stat-usa.gov/ says the
average time a company is in business in the US if a year of
insolvency is defined as the company going "bust"? (This is the DOC's
definition, by the way.) Not surprising, using 1 / sqrt (t), where t
is in years provides some insight. (The metrics are surprising close
to the predicted value of 4 years-3.8 years.)

Interesting, huh? The reason these things all predict out fairly well
is that fractals are self-similar, (ie., the same statistics hold
true, irregardless of scale. Define the scale, such as one bad year
meaning "bust," and the probabilities will all scale
proportionally. In fractional Brownian systems-like equity prices,
industrial markets, and corporate P&L's-the probabilities will always
be some kind of square or square root law. These types of
probabilities are called power laws-surprising name for them-and are
characteristic of fractals-by definition.)

Want some popular misconceptions of fractals ie., questions often ask
by business people?

    Q1: Why go to all the trouble of trying to run a company if its
    destiny is out of my hands, ie., determined by power laws.

    A1: If you don't play, you can't win. But playing is not a
    guarantee that you will win-only a necessary, but insufficient,
    requirement. Power law probabilities mean, however, that luck
    plays a significant role in success, ie., it would probably be
    better to consider business as gambling as opposed to a
    deterministic science, (no insults to Harvard'isms intended.)

    Q2: Most probabilities in power law things have a 1 / sqrt (time)
    type of scenario, which is 50% when t = 4. This means that most of
    the time, things will happen when t = 4, right?

    A2: No. It is a popular misconception derived from an implied
    meaning of Gaussian distributions, (the central limit theorem, to
    be exact,) that the mean represents "most things." In these types
    of distributions, however, this is not true. What 1 / sqrt (t)
    means is that half the things will have occurred by t = 4, and
    half will not have. Most things occur when t = 1. What we have
    here is a distribution where the "average" is not the "mean," and
    does not represent "most" things.

    Q3: Then, does 1 / sqrt (t) mean that there is a predictable
    mechanism that can be exploited in business?

    A3: Yes, and no. It does not mean that the chances of an up or
    down movement in something is any greater because t < 4. The
    chances remain 50%/50%. However, it does mean that extended
    duration excursions from "average" are to be expected-and this can
    be exploited in a probabilistic nature.

    Q4: Can't fractal metrics be used against a CEO by the BOD as a
    justification for termination?

    A4: Yes, at least in principle. But there are, however, pragmatic
    issues in doing so. For the typical company in the US, metrics
    would have to be taken, by the day, on the CEO's pro forma for
    13,000 days (about 36 years!) for the BOD to be 49% confident that
    a company's bad fortune was the CEO's fault. For 50% confidence,
    it would take all of eternity. (It is not trivial to calculate
    these values, and depends, greatly, on how fast the industrial
    market is growing with respect to how much market share the
    company is loosing. In a high growth market-ie., one that is
    growing at 20% per year-a 50% confidence would require only a
    little over 5 years of annual pro forma. Bear in mind that a 50%
    confidence level is not very good, and probably wouldn't stand in
    a court of law.)  The reason for these kinds of long durations are
    the same as mentioned in A3, ie., what is the chances that it was
    not the fault of the CEO, but just fate that caused the company's
    bad fortune for such an extended duration. (It is an important
    concept that, although these probabilities move to 50% in just 4
    time units, they are very "sluggish" after that because the
    "tails" of the 1 / sqrt (t) function drop off very slowly.) The
    converse is also sobering. The chances that a CEO can have great
    success, (when it was not justified by capabilities,) for 4 years
    is 50%.

    Q5: What is the fastest, sustainable, growth rate for a company?

    A5: In theory, a factor of two in compound annual growth rate is
    sustainable. But only under theoretically idealized conditions.

    Q6: How far can we see into the future in commerce?

    A6: This is a complicated question. The present is determined, on
    the average, by no more than 4 time units in the past, and the
    future is forecastable for no more than 4 time units.  The
    question is, what's a time unit? If you are looking at daily
    numbers, (like equity indices, or daily operational numbers,) then
    any forecast that is based on daily data is good for, on the
    average, 4 days. If you are looking at annual market numbers for
    an industry, then it is 4 years. That is the way fractal things
    work. They have a "horizon of visibility" at all time scales, be
    it minutes, days, weeks, years, or decades, beyond which, nothing
    can be known with better than 50% accuracy, (ie., a flip of the
    coin is as accurate a forecasting mechanism as any.)  It is
    inappropriate to attempt to use data at one time scale to forecast
    another time scale-ie., using daily numbers to forecast annual
    outcomes is impossible. (It is not only possible, but frequently
    the case, that data at small time scales show increasing trends,
    but at large time scales, shows decreasing trends-and vice versa.)

    There is an interesting corollary to this concept. As you move
    down into an organization, not only should there be a "finer
    granularity" of management and operations, but the time scale used
    should also be shorter.  Managing the pro forma for a year through
    managing day to day operations is an invalid concept. Either have
    people manage the day to day operations, or the annual operations,
    but not both. (The statistics scale on the square root of the time
    scale-increasing the time scale by a factor of two, scales the
    statistics by a factor of 1.4. It is difficult to change "mind
    sets" when jumping time scales-for example, business variances-as
    per MBA-is different at each level in the organization.)

    Q7: So, do fractal concepts mean that there can never be a science
    of commerce?

    A7: In the traditional sense of deterministic science, (like
    classical physics, for example,) that is exactly what it means. It
    means that no perfect solution to commerce exists-we will have to
    use an "approximately good" solution, which will sometimes work
    out, and sometimes not. (That is why the answer to Q1 mentioned
    the word "gambling.")

    Q8: So, when management consultants claim that they have solutions
    to business problems they are trying to sell "snake oil?"

    A8: Although many management consultants truly believe in their
    concepts, very few that purport scientific merit stand up to
    formal scrutiny. Management consultants usually work with
    companies that have been "down" for more than 4 years, and almost
    never with one that has been "down" for less than 4. (This is a
    widely known empirical observation in the consulting industry. The
    statement is usually that it takes about four years for the
    management of the company to become pliable to outside
    intervention. It is, inadvertently, exploiting the probabilities
    of the 1 / sqrt (t) phenomena.)

    Q9: What is the optimal fraction of my product portfolio that
    should be industry standard products, (in relation to proprietary
    products)?

    A9: Theoretically, one standard deviation, (ie., root mean square,
    or 68%,) of your gross revenue should come from industry standard
    products, (in the long run,) which is optimal.

    Q10: How do you forecast product life cycles?

    A10: By dynamic estimate. If you are running operations by the
    month, then the probability of a product's life continuing into
    the next month is 1 / sqrt (t), where t is the number of months
    since product introduction. Additionally, once product shipments
    have decreased for four months, the product life cycle is,
    "effectively," over. Meaning that eight months is a good
    conceptual estimate for product life cycles that are managed by
    the month. Using dynamic estimation by probability, as calculated
    by 1 / sqrt (t), however, is highly recommended, since, by
    lottery, one might get a product that continues into the
    "sluggish" tails of the 1 / sqrt (t) function. There is a 50%
    chance of that happening, (in case you are a gambler, and want to
    set up your budget forecasts,) and then you switch time scales,
    (to by-the-year,) by scaling the statistics from the by-the-month
    data by a factor 1 / sqrt (12).

    Q11: How do I know if a product is a flop?

    A11: If you are managing products by the month, then if a product
    has not made a sustained four month growth after introduction,
    there is only a 50% chance it will ever succeed.

    Q12: How do I know if a product is a success?

    A12: If you are managing products in an industry that manages
    products by the month, and a product has a four month lead on the
    rest of the industry, chances are 50% that the product will,
    eventually, dominate the industry, (says something about being
    first in niche markets, huh?) and continue on into a market lock.
    Many theoreticians, (and empiricists, also,) are of the opinion
    that exploitation of the 50/50 lottery in niche markets is what
    high technology business is all about, and that the best strategy
    is to be first in the market, at all costs.

    Q13: Is there a science of management?

    A13: Not in the traditional sense of deterministic science, (like
    classical physics, for example,) and not in the sense of the
    statistical sciences, (like analysis by demographics and/or
    correlation studies.) But there is in the sense of the
    probabilistic sciences, like the quantum mechanics of modern
    physics.

    There is an interesting interpretation related to this
    statement. The objective of management is to maximize the ratio of
    the average of the marginal increments of something, (like the
    P&L, or market share, for example,) to the root mean square of the
    marginal increments, while, simultaneously, equating the average
    to the square of the root mean square of the increments. Sounds
    like double talk, huh? Not really. What I said would maximize
    growth, while simultaneously, minimizing the risk to the
    growth. How does one do that? There are three ways: 1) increase
    the average, 2) decrease the root mean square, 3) work the issues
    of the ratios of the two. The first two alternatives are the
    traditional methods used in business. Working the first
    alternative will result in phenomenal short term growth, for a
    while, followed by a "crash" and the company going "bust." Working
    the second alternative will result in eternally "stilted" growth
    and inevitable "bust," but without the "crash."  Working the third
    alternative is interesting, because this ratio is a metric of the
    effectiveness of management decision process, (specifically, this
    ratio is the likelihood that something that operates in a
    probabilistic fashion, like a P&L, will increase.) Note that there
    are two variables that interrelate to each other and which can be
    manipulated in an optimal fashion.  So, how does one do that?
    Hint: management decision probabilities add root mean square,
    meaning that the more people's perspective that is included in the
    decision process, the better. Better by the square root of the
    number of people/perspectives, as a matter of fact. And, how do
    you get all those people to agree and formulate a decision? I
    don't know, but I can tell you how to measure how well they did
    it. That is the average of the marginal increments. So, there is a
    metric that will tell managers what needs to be done, (but not how
    to do it.)  Consensus issues control the average of the marginal
    increments, and knowledge/perspective issues control the root mean
    square. The ratio controls the growth in P&L, or whatever. And,
    not only that, there is an optimal growth, with minimal risk.

    Q14: How long before a business model is obsolete?

    A14: For executives running a company on an annual basis, it is
    four years, on average. However, the probability of a business
    model lasting longer falls off very slowly-for example, the
    probability of a business model lasting a decade is about
    32%. That's the good news. The bad news is that 29% of the
    business models will be obsolete within the current year. Point?
    It means that the executive staff of a company will have to have
    the capability to reinvent itself, and its concepts,
    quickly. Without falling apart.

    Q15: How long do I have to get into a market?

    A15: The market will grow at sqrt (t), on the average. The
    duration of the market will be proportional to 1 / sqrt (t), on
    average, (actually, 1 / sqrt (t) is an approximation to erf (1 /
    sqrt (t)) for large t.)  The product of the two is the average
    gross revenue to be made from entering industrial markets. Bottom
    line? Suppose you are in a market where the competitors manage by
    the month. If you are more than 5 or 6 months late, someone else
    has made all the money. (The theoretical number is that at 5
    months, 95% of the gross revenue from the market is history.)
    Point? Corporate agility is important. (Another of economics most
    profound insights-history waits for no one-he who does the most
    the quickest wins-the meek will not inherit the earth, the agile
    will, etc.) Another point? It means that the executive staff of a
    company will have to manage change, quickly. Without falling
    apart.

    Q16: Do business cycles exist?

    A16: Yes, and no, (which is a nice hedged statement.) If you
    measure business activity, using an annual time scale, over, say,
    several decades, you will see the "cyclic" phenomena of the 1 /
    sqrt (t) = 4 years assert itself, (ie., a 50% chance of change
    every four years, or so.) However, using many decades, the "tails"
    of the 1 / sqrt (t) distribution will assert their selves, and the
    "cyclic" phenomena will disappear. (A more precise probability for
    erf (1 / sqrt (t)) is 4.4 years, which is very close to the 5 year
    business cycles cited in business journals.)

    Q17: What's the difference between 1 / sqrt (t) and erf (1 / sqrt
    (t))?

    A17: I'm trying to get executives to think in probabilistic terms,
    (ie., be good gamblers.) 1 / sqrt (t) is a "conceptual"
    approximation to many of the probabilities involved in commerce.
    It is a "reasonable" approximation to erf (1 / sqrt (t)), at least
    for large t. The term erf (x / sqrt (2)) is the error function
    associated with the normal, (ie., Gaussian bell,) curve. So, if
    more precision is desired, the error function can be computed with
    scientific calculators, spread sheets, or standard math tables.
    For example, to evaluate erf (2.3), one proceeds as follows: Since
    x / sqrt (2) is 2.3, one finds x = 2.3 * sqrt (2) = 3.25. The
    value of the normal distribution for 3.25 standard deviations is
    0.9994. Subtracting 0.5 from this value, one has 0.4994. Thus, erf
    (2.3) = 2 * 0.4994 = 0.9988. For "conceptual" arguments, such
    precision is seldom justified.

    Q18: What are the chances of a decision to develop a product
    becoming obsolete before the product is developed? I mean, if it
    takes 8 months to develop a product, what are the chances that the
    product will still be viable in the marketplace 8 months from now?

    A18: Assuming that you are managing things on a monthly basis, and
    that is commensurate with the industry, 1 / sqrt (8) = 35.4%.

    Q19: How is my revenue distributed across my products?

    A19: 84% of your gross revenue, in the long run, on the average,
    will be generated by 16% of your products. Point? Be careful how
    you trim your product line. This is an assertive
    probability. Trimming your losers could result in you getting 84%
    of your gross revenue of a smaller number. Best bet? Figure out a
    way of supporting your losers as economically as possible-ie.,
    minimal cost.

    Q20: How much does product diversification enhance my chances
    of corporate survival?

    A20: If your executives can manage many products effectively, then
    the chances of your company going "bust" is reduced by 1 / sqrt
    (n), where n is the number of products in your product portfolio.
    (Case in point? General Electric, who has watched half of the
    companies that were ever listed on the NYSE come, and go. A lot of
    those companies had good ideas, too. For a while. )

--

John Conover, john@email.johncon.com, http://www.johncon.com/


Copyright © 1998 John Conover, john@email.johncon.com. All Rights Reserved.
Last modified: Fri Mar 26 18:54:32 PST 1999 $Id: 980130234517.26676.html,v 1.0 2001/11/17 23:05:50 conover Exp $
Valid HTML 4.0!