Stock Trading Evolution


E

Elle

Below is some interesting commentary that, among other
things, tends to reinforce the notion that as soon as a new
investing strategy appears, its effectiveness quickly
diminishes, due to rapid exploitation of the strategy and
market action. In other words, if there is an investing holy
grail, its lifetime is short.

--- "A Smarter Computer to Pick Stocks," NY Times, Nov
24 ---
[Excerpt]
Studies estimate that a third of all stock trades in the
United States were driven by automatic algorithms last year,
contributing to an explosion in stock market activity.
Between 1995 and 2005, the average daily volume of shares
traded on the New York Stock Exchange increased to 1.6
billion from 346 million.

But in recent years, as algorithms and traditional
quantitative techniques have multiplied, their successes
have slowed.

"Now it's an arms race," said Andrew Lo, director of the
Massachusetts Institute of Technology's Laboratory for
Financial Engineering. "Everyone is building more
sophisticated algorithms, and the more competition exists,
the smaller the profits."
 
Ad

Advertisements

J

jose.bailen

Computers are key but they are just a tool, a very sophisticated tool.
In my case, to pick my stocks, I use an econometric package and an
error correction model to estimated future earnings of a given firm. An
error correction model is a non-linear, non-stationary model which fits
very well with past data, and it can give a relatively accurate
projection of future earnings. For a value investor, the estimation of
future earnings is the most important variable to pick -or not- a
stock, since discounted future earnings gives the intrinsic value of
the stock, which is compared to market value.
 
E

Elle

For a value investor, the estimation of
future earnings is the most important variable to pick -or
not- a
stock, since discounted future earnings gives the
intrinsic value of
the stock, which is compared to market value.
Jose, I think folks should take the above as an opinion
only. For one thing, Ben Graham, the so-called father of
value investing, does not recommend relying on future
earnings. Such calculations of the future demand too many
assumptions.

My sense is that the algorithms about which the article
talks seem to use a mix of conventional wisdom for choosing
stocks; some "technical" analysis; and information of all
flavors and varying specificity, depending on the industry
and stock, concerning economic conditions.

Ben Graham has a funny line about "technical analysis" and
how there's nothing technical about such methodologies. I
agree with him (big deal; little Elle agrees with the old
sage, I know), hence my quotation marks, as well as my
'raised eyebrow' at what some of these algorithms claim to
do.
 
B

beliavsky

Computers are key but they are just a tool, a very sophisticated tool.
In my case, to pick my stocks, I use an econometric package and an
error correction model to estimated future earnings of a given firm. An
error correction model is a non-linear, non-stationary model which fits
very well with past data, and it can give a relatively accurate
projection of future earnings.
What is the cointegrating vector -- what equilibrium relationship does
your model assume?
For a value investor, the estimation of
future earnings is the most important variable to pick -or not- a
stock, since discounted future earnings gives the intrinsic value of
the stock, which is compared to market value.
It is discounted DIVIDENDS, not EARNINGS, that equal the fair value of
the stock.
 
J

jose.bailen

Dividends depend on the company's policy with respect to their
earnings, they can either distribute them -through dividends- or
reinvest them and then increase the value of the stock. For example,
Microsoft did not distribute any dividends until very recently (2001 I
believe), yet the company was worth in 1986 much more than whatever
dividends have been distributing since 2001 or so. So it is discounted
EARNINGS, not discounted dividends what determine the value of an
enterprise.

With respect to the technicalities of the estimation of an error
correction model -unit roots, the cointegrating vector, etc...- they
are a bit too sophisticated for this forum. I have a paper on the
Spanish housing market which uses the same analysis and methodology
used for stocks, you may see it here:
http://www.um.es/analisiseco/documentos/Jose-Bailen.pdf
 
J

jose.bailen

You need to have some estimate of future earnings if you want to know
how much a company is worth. I agree that it is tricky to do these
estimates, this is why I developed my own model -an error correction
model- which is based on the past record of earnings and reasonable
medium term assumptions (5-year) assumptions on the evolution of
exogenous variables, such as industry growth (a proxy of growth of
sales), and the relative prices of the company/industry good with
respect to salaries and interest rates. These exogenous variables,
given the parameters derived from the historical relationship, give me
a non-linear dynamic model to estimate profits for the next five years.
For the long term, I assume that profits will increase at the same rate
as its historical average of 6.1 percent (a rather conservative
assumption since my stocks had an above-average performance in the
previous 10 year period, and they are micro and small caps with a
higher growth potential than the market average). I estimate the model
using this econometric package: http://www.oxmetrics.net/ . I learned
how to use it in a couple of courses given by its developer, professor
Hendry of Oxford University, at the IMF.


As you may appreciate, I picked my stocks by using something more than
just a stock screener...
 
Ad

Advertisements

E

Elle

Backtest it, tell the group why Hendry is not filthy rich at
this point, and report back. ;-)
 
J

jose.bailen

Hendry is an econometrician, a university professor, I don't think he
has ever applied his econometric package to finance (he applied it to a
model of money demand for the UK and housing market, also for the UK).
To my knowledge, I'm not sure if anyone has ever applied error
correction models to the enterprise income estimates and thus the
calculation of the intrinsic value of an enterprise. The technique is
relatively new -it has been around for about 10 years-, and I don't
know if any mutual fund has applied it yet. I like this model because
of the non stationarity and non linearity of the net income serie, this
makes an error correction model an ideal technique to get these
estimates. Of course, as every other econometric model, if there is
structural change/location shifts, the projections are meaningless. But
given this caveat, I think it is the best we can do given the data we
have, from the empirical point of view.
 
T

TB

Elle said:
Below is some interesting commentary that, among other
things, tends to reinforce the notion that as soon as a new
investing strategy appears, its effectiveness quickly
diminishes
[Excerpt]
Studies estimate that a third of all stock trades in the
United States were driven by automatic algorithms last year,
contributing to an explosion in stock market activity.
I believe that premise, that most mispricing should be temporary. Though
a few anomalies (such as the returns of value stocks) test that assumption.

But WRT program trading - my understanding is that a lot of that
activity isn't the type of trading you might be thinking of. I think a
substantial portion of the volume is more mundane arbitrage trading,
including ETF arbitrage. For example if an S&P 500 ETF is priced a bit
too high relative to the value of its component securities, even after
factoring in transaction costs, you short the ETF and purchase the
basket of securities, producing a risk-free profit. There are similar
trades for other products and derivatives. If computers weren't working
these trades ETFs would face the pricing problems of closed-end funds.

The growth in ETFs means ETF-arbitrage program trading should steadily
increase, and those profits are always going to be there for the firms
with the resources to quickly identify and trade around them. I imagine
too many firms may get into it for it to be profitable enough, but that
should self-correct.

-Tad
 
T

TB

You need to have some estimate of future earnings if you want to know
how much a company is worth. I agree that it is tricky to do these
estimates, this is why I developed my own model -an error correction
model- which is based on the past record of earnings and reasonable
medium term assumptions (5-year) assumptions on the evolution of
exogenous variables, such as industry growth (a proxy of growth of
sales), and the relative prices of the company/industry good with
respect to salaries and interest rates.
Jose, good for you for developing a model like that, and very
interesting paper on housing (which could be a thread in itself...did
you model the US market as well?). [RE: the equity model - My personal
belief is that earnings predictions even a couple years out are
difficult/impossible to make any better than the market does. The best
we can do is watch for overreactions to earnings misses or other bad
news, and adopt a longer-term attitude than the typical institutional
investor.]

I think you & Beliavsky can find a common ground this way...if you want
to value a stock, and aren't assuming infinite life, a dividend-discount
model should include a "residual value" or "liquidation value" or
"buyout value" term at the end of your stream of projected dividends.
And of course, you'd need to look to earnings relative to dividends paid
to estimate retained earnings and the value of the enterprise at that
end point. I don't see this as strictly a theoretical exercise because
"going private" with fat cash flow & a solid balance sheet, or being
acquired, is an common exit for businesses that produce the strong cash
flow that could be, but isn't necessarily, used for dividends.

This wraps into the other Elle thread, on the importance of dividends in
choosing an investment. I think for a lot of reasons, you can't focus on
dividends when selecting stocks, because there are too many things that
get in the way between earnings and dividend policy. Unless you look at
total return you're ignoring external factors that might be affecting
dividend policy (eg for a REIT, tax constraints; for a stock,
acquisition plans).

-Tad
 
P

Paul Michael Brown

I use an econometric package and an error correction model
What is the cointegrating vector -- what equilibrium
relationship does your model assume?
As they say down South -- Say what again?

This is why I am a Random Walk, Efficient Market, John Bogle, Index Fund
kinda investor. ;-)

But that doesn't mean I don't enjoy reading posts from those of you more
academically and mathematically inclined. The group never ceases to be a
learning experience for me.
 
Ad

Advertisements

J

Jose Bailen

An error-correction model is a dynamic model in which the movement of
the variables in any periods is related to the previous period's gap
from long-run equilibrium. What you do is, first, estimate the long
term equilibrium relation between the variables (in this case, earnings
of a given company -endogenous variable- and the exogenous variables:
company's industry sales, relative price of the good sold by the
company/industry; and salaries, and interest rates (which explains most
of the cost of goods sold by the company). These exogenous variables
-which are non-stationary, i.e, they have a trend to increase (or
decrease) over time- need to be cointegrated with the price of the
stocks, i.e, a linear correlation of earnings and -say- salaries paid
by the industry needs to be stationary. Intuitively, they should move
in the same direction over the long term.

The error correction model provides both the long term equilibrium and
the short term dynamics. For the model's forecast, you need to make
some assumption on the evolution of the exogenous variables. Once you
make these assumptions, you obtain a forecast of the endogenous
variable -in this case, earnings- and therefore you obtain the
intrinsic value of the company, and compare it with its actual market
value.
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Top