ExtremeHurst For Bloomberg Guide

Users Guide for ExtremeHurst™
on Bloomberg
Parallax Financial Research

6/27/2014

Users Guide for ExtremeHurst™ on Bloomberg

 What is ExtremeHurst? Page 2
 How do I use ExtremeHurst? Page 7
 The Application APPS EHURST <Go> Page 9
o Security & Filter Selection Page 10
o Signal Search and Progress Page 13
o ExtremeHurst Signal List Page 14
o ExtremeHurst Output Options Page 19
 The Custom Studies APPS CS:PFR <Go> Page 24
o ExtremeHurst
o Price‐Volume Crossovers Page 26
o SmartChannel Page 28
 ExtremeHurst Science Page 30
 ExtremeHurst Testing and Statistics Page 35
 ExtremeHurst Users Page 44
Parallax Financial Research 1

What is ExtremeHurst?
ExtremeHurst is a quantitative detector of extreme investor behavior that signals the beginning or end
of a trend. Strong trend‐persistent stock price movements are evidence of positive feedback (i.e.,
investors buying because the price is rising, driving prices higher), while extremes of mean reversion are
evidence of negative feedback. Extremes of both trend persistency and mean reversion are quantified
via multiple measurements of the Hurst exponent. Parallax found that Hurst extremes coupled with
other characteristics, signal the beginning or end of market trends. ExtremeHurst signals are fully
characterized by the presence of discrete scale invariance, accelerating price, log‐periodic cycles, and
volume anomalies. This App allows the user to search global markets for signals on intraday, daily,
weekly, monthly, and quarterly time scales. Signals have been found to persist for up to 20 bars.
ExtremeHurst exploits the science of non‐linear dynamics to identify unique and predictive signals
occurring in freely trading auction markets. There are two types of ExtremeHurst signals that we call
“Extensions” and “Compressions”. These correspond to the extreme high and low ends of the Hurst
exponent distribution.
Figure 1. This is an example of a weekly scale Extension top signal.


Figure 2. Histogram of Hurst exponent measurements taken from the S&P 500
The Hurst exponent, as applied to a financial series, represents the degree of randomness which is
present. Deviations from random take the form of mean‐reversion or trend persistency. The picture
below shows five time series with different Hurst exponents. The topmost example is the most mean‐
reverting, while the bottom example shows the most trend persistency. The middle one is random.
Figure 3. Time series generated with different Hurst exponents

Extensions correspond to extreme levels of trend persistency on multiple scales, and mark the end of
trend persistent periods, either at market tops or bottoms. The picture below shows multiple Hurst

exponent measurements for a particular security at different scales. The shortest scales are on the
bottom. Red represents persistent down trends. Green represents persistent up trends. Gray shows
random price movement, and blue represents mean‐reverting exponents. Note the column of red at
the price low.
Figure 4. Color-coded Hurst exponents at multiple scales (graph not available on BB)
Many factors influence investors, but we choose to focus on the extremes of competitive or cooperative
behavior from a macro perspective. Feedback is at the heart of why ExtremeHurst works. Investors
sometimes behave as herds, selling because the price is dropping, or buying because it’s going up.
Panics and manias are large scale examples of this. However, when most investors agree that a stock is
going up or down, they’ve probably already acted on their belief, and the buying or selling dries up,
leading to a reversal. Panic Mania

Figure 5. Extreme investor behavior.
Because these events are scientifically similar to critical failure points in materials, log periodic
“foreshocks” are visible preceding most signals, as well as mirror image “aftershocks” following signals.

Aftershocks are also log‐periodic, but are expanding, which means the highs and lows are getting farther
apart logarithmically over time. The picture below is a simulated example:
Figure 6. Simulated log-periodic oscillations culminating in a critical reversal point and then
expanding away from the critical point (Top Extension)


Compressions correspond to extreme levels of mean‐reversion, or investor competition, on multiple
scales. These signals mark the beginning of trends by finding when the vigorous competition between
supply and demand has reached a critical point. Price is expected to move very rapidly away from its
current price following a compression.
Figure 7. Competition between supply and demand reaches a crescendo at compression signals
Another way to think of these signals is to envision a mass on a spring. When the spring is compressed,
the system has high potential energy. When the mass is released, it will move very fast away from its
starting position. In fact, it will then move too far. An extension occurs as the spring reaches its full
extent, just before it settles back:
Compression Extension
Figure 8. A harmonic oscillator is a great way to envision a compression and extension

How do I use ExtremeHurst?
The ExtremeHurst signals work on all freely traded securities and on all time scales, provided sufficient
liquidity is present. For a “crowd effect” to occur, a crowd must be present.
Figure 9. Examples of each ExtremeHurst signal type
The predictive edge lasts 15‐20 price “bars”, so for monthly scale data, this means the effect might
continue for as long as 20 months.
The Extension signals predict retracement or sideways periods, so redeployment of capital may be the
best strategic move. Aggressive speculators like to play the log‐periodic cycles, so we have included the
expected top and bottom timing marks on the charts and in the signal file.
Compressions mark the start of new trends, but we never know the direction, so a straddle or triggered
L/S entry is required.
Signals have at least four important uses:
1. Trade positioning
2. Profit taking.
3. Identifying transitions from a trending to a sideways market or from a sideways to a trending
market.
4. When the crowd runs a security too far up or down, prices can deviate significantly from a
reasonable valuation and diminish portfolio performance. Having an idea about when these
effects are occurring can be very valuable.
Signals are ranked from 1 to 100 depending on how closely they correspond to the ideal signal

An ideal signal exhibits four characteristics:
1. A price acceleration that deviates significantly from normal Gaussian expectations
2. Extreme Hurst Exponent measurements on multiple scales
3. Log‐periodic price ripples which converge to the signal date
4. Unique price and volume sequencing
We combine factors using a pre‐trained neural net to produce our final signal rank
We have created the ExtremeHurst App on Bloomberg to enable users to search world markets for
these predictive signals on intraday, daily, weekly, monthly, and even quarterly time scales.

The Application (on Bloomberg at APPS HURST <Go>)

ExtremeHurst ExtremeHurst is a single Windows dialog box. It is divided into six sections from
App Window top to bottom that correspond to the workflow sequence. These sections include
security and filter selection, the search progress bars, the signal list, the chart
area, the zoom control, and the output options.

CPU Usage ExtremeHurst performs millions of complex mathematical calculations including
Fast Fourier transforms, neural network math, and trigonometry calls during each
run. You will notice a sharp increase in CPU usage during searches and on
occasion the processing may even appear to halt. Be patient, it will complete.
Bloomberg Data Bloomberg clients have monthly data usage limitations. We recommend that
Limits you carefully select a security universe, search once after the close each day for
daily signals, once a week after the weekly close for weekly scale signals, etc.,
so you stay within your limits. ExtremeHurst reads 129 historical bars worth of
OHLCV data for each security that passes through the filters.

Security & Filter Selection
Let’s take a closer look at security and filter selection: The user can select which securities to search by
either reading in a text file with one security name per line, retrieving each security holding from an
index (SPX for S&P 500, NKY for Nikkei, CAC, DAX, etc.), reading securities from a Bloomberg client
portfolio (PRTU <Go>), or by selecting a region or country to access all their respective securities. The
next step is to select the type of security, and time scale desired. Keep in mind that our signals remain
active for an average of 20 bars. This means that a single quarterly signal may be predictive for 5 years,
while a daily scale event lasts for a month or less. The next menu has a list of filters to apply. The
volume filter ensures that securities with volume actually have volume on every trading day. It also
ensures a sufficient level of volume for reasonable investment. The data filter guards against spikes,
excessive gaps, missing data, or improperly sequenced data. The price filter sets a minimum price level.
Checking “New Signal” results in signals that have just occurred; while unchecked it results in new AND
older signals that are still active. By checking ”Remember Filtered”, the program will remember which
securities were excluded and not search them again. Three signal quality thresholds are provided. We
have found that some researchers don’t mind seeing less predictive signals, while others just want the
few best ones. We recommend leaving this on “High”.

Single Security Enter any single security
Securities List A text file created using NotePad or other text editor with a list of securities, one
per line, in the following format: <security name> space <exchange code> space
<yellow‐key>. For example:

Index Holdings Search just the securities that make up an index by typing the Bloomberg index
name at the prompt. For example, SPX for the S&P 500 index:
Client Portfolio Bloomberg offers clients the ability to keep portfolios on their system using PRTU
<go>. If you wish to search only your portfolio securities then enter your portfolio
name at the prompt. It should be in the following format: U1234567‐1
Country Search all securities by country.
Region Search all securities by region.

Security Select which type or types of securities to search for ExtremeHurst signals


Time Scale Select which time scale to search for ExtremeHurst signals

Liquidity Filter With this option checked, the average daily dollar trading volume for Indices and
equities must exceed a minimum limit to be considered.
Data Filter Check this option to filter out securities with missing data, recent data spikes,
inactive bars, insufficient history, excessive gapping, or improper sequencing
Price Filter This option limits the minimum allowable price of an equity or index
New Signals Check this option to search for new signals only. If it is unchecked the search will
find all signals that are still active….but it will take longer.
Remember Check this option for the program to remember all the securities that have been
Filtered disqualified for signal search. This memory will persist even when the program
is closed
Preferences Certain combinations of search options can enable users to have either more
signals (at the expense of compute time), better signals by applying still more
stringent filters, or faster calculations (at the expense of getting fewer signals).
Best Quality Select this option to find only signals that have the highest possible quality (90+)
High Quality This is our recommended setting for finding signals (80+)
Medium Quality Some users like to see lots of signals even if they have a lower overall quality.
(70+)
Explore Quality These are low quality at 50+ but have been requested by some users. Not
recommended for trading


Signal Search and Progress

Search for Signals Press the search button to download historical price data from Bloomberg for the
selected securities and then search that data for ExtremeHurst signals. The
download procedure first checks that the average daily dollar volume is greater
than the limit before requesting additional data. Look at the “Filtered” section
below the STOP button to find out how many securities passed all the filters and
have finished being run through the ExtremeHurst signal processor. In the
example above, we show the search section before and during a search of S&P
500 stocks. Note that the search button is grayed‐out until all downloading and
calculations are complete. As shown above, out of 275 securities downloaded,
228 had been passed to the processor, and 27 had completed so far.
STOP The search procedure gathers data in chunks and then processes that data before
gathering the next chunk. Press STOP to break the cycle and stop the processor
from gathering additional data. It will complete the signal search using the data
already downloaded however.
CPU Usage ExtremeHurst performs millions of complex mathematical calculations including
Fast Fourier transforms, neural network math, and trigonometry calls during each
run. You will notice a sharp increase in CPU usage during searches and on
occasion the processing may even appear to halt. Be patient, it will complete.
Data Usage Bloomberg clients have monthly data usage limitations. We recommend that you
carefully select a security universe, search for “New Signals” only, and then search
once after the close each day for daily signals, once a week after the weekly close
for weekly signals, etc., so you stay
within your limits. ExtremeHurst
reads 129 historical bars worth of
OHLCV data for each security that
passes through the filters. If you
uncheck the New Signals button the
application will take longer to run
and require 15% more data. If you hit your limits you’ll see:

Data Warning If you ask for more than 1000
securities then this message comes
up. As an example, In the US, the
almost 20,000 listed equities are
cut down to less than 2,000 by the
filters. If you are unsure, press the
STOP button part way through, or
just make a security list in a text
file and read it in.
ExtremeHurst Signal List
Signals are listed in the window as the search progresses. The list has one signal per line and is
described by its symbol name, exchange, security type, time scale, signal type, quality rank, forecast,
and active dates. The button marked “More” is used to find historical signals and do a statistical test on
them. A yellow box indicates that insufficient data was present to do this test. Green means the test
was significant, while red means the opposite. Use the scroll bars to move through the signal list.

Symbol The Bloomberg symbol for this security
Exch The Bloomberg composite exchange where the security is listed
Type Abbreviation for the Bloomberg yellow key. EQ=Equity, CM=Commodity,
CR=Currency, IN=Index, ET=Exchange traded fund
Scale Abbreviation for the sampling frequency, DY=Daily, WK=Weekly, MO=Monthly,
and QT=Quarterly

Signal There are two types of ExtremeHurst
signals that we call “Extensions” and
“Compressions”. These correspond to
the extreme high and low ends of the
Hurst Exponent distribution. There are
two types of Extensions, top and
bottom. Extensions mark the end of
trends on that scale. Compressions are
points of high potential energy that
occur just before a new trend erupts.
Rank Rank varies from 0 to 100 corresponding to the signal quality. To be a high quality
signal, extreme high or low consistent Hurst Exponents, log‐periodic oscillations
(Sornette), and expected volume behavior need to be easily distinguishable once
data corrections and filters are applied. The higher the rank, the better the
signals expected outcome.
Forecast Abbreviation for the behavior we expect from the three signal types.
Valid From Date from which the forecasted behavior should start to appear.
Until This is the Date after which it is unlikely that signal effects will be evident. Note
here that this is set at 20 bars currently. Expect signals that had longer build‐up
times to have effects lasting longer, and vice versa.

You can also right click on each line to bring up a convenient Bloomberg terminal task menu:


GP: Display Bloomberg Chart

DES: Security Description

Chart of Selected Signal
Marker Symbol Float over

Extension Top marker

Compression marker

Extension Bottom marker

Price bars

Marker Symbol Float over

Extension Cycle Top markers

or

Extension Cycle Bottom markers

or

Accelerating Trend
Price

Log‐Periodic Oscillations
Price

Average Compression trend path

or

Price Volume Cross Sell

Price Volume Cross Buy


ExtremeHurst Output Options

There are seven output‐related buttons across the bottom. “Save Signals” allows the user to write the
signals to a CSV file. The “Copy Signals” button places the same signal data in the clipboard, so it can be
pasted in a spreadsheet or other compatible document. “Clear Signals” wipes the signal list clean. The
Print Screen” button sends an image of the current dialog box to the printer. It does not save all the
signals. “Copy Screen” sends an image of the current dialog box to the clipboard. The “Add History”
adds historical signals and other studies to the chart. Finally, the “ChartBook+” button accumulates the
user’s favorite chart images for later printing and viewing.
Press this button to save the signal list to a comma delimited text file.
Press this button to save the signal list to the Windows clipboard. Below is the signal list
pasted into EXCEL.

Press this button to clear the signal list and chart.
Press this button to print a copy of the screen image.


Press this button to copy an image of the screen to the windows clipboard. This allows you
to paste the image in a report or email.


Press the “History+” button to add historical signals and additional studies to each chart.
Once pressed, we gather additional data for the particular security and do a statistical calculation to see
if other ExtremeHurst signals in the past behaved as expected. We also add two other signal types,
SmartChannel (Custom Study “PFSC”) and Price‐Volume Crossovers (Custom Study “PFPVC”). The Price‐
Volume crossover signals are useful bottom and top indications that last for about up to 6 bars.
Press this button to accumulate the current chart image to a book for later printing and
review.

Our Help screen is shown below. Our links were not allowed to be live.

Parallax Financial Research, Inc. is a small scientific research boutique in Redmond, WA. The firm
empowers professional money managers with a distinct edge through its unique fundamental and
quantitative stock and commodity models. Our models are composed of individual "predictors" which
are based on the financial application of both chaos and complexity theories, and presented in clear
visualizations. These predictors are blended into forecasting models using genetically‐enhanced neural
networks. Our careful application of these mathematical modeling techniques yields unique and
powerful solutions to enhance manager performance. Visit our website at www.pfr.com


The Custom Studies (APPS CS: PFR <Go>)
The ExtremeHurst™ custom study (“PFEH”) allows the user to overlay our study on a Bloomberg chart.
Once in the Bloomberg charting screen, it is easy to change time scales, look at long signal histories, or
follow ExtremeHurst in real time.

There are a few settings associated with our custom study: The signal quality varies between medium
and superior. Medium quality signals have ranks of 70‐80, High quality signals have ranks from 80‐90,
and Superior signals have ranks of 90‐100. It is often useful to examine charts using a medium minimum
signal quality so that more signals are visible. The Historical Signals setting allows you to speed up the
study by choosing to just display the most recent signal.

ExtremeHurst signal commentary can be accessed using the Bloomberg Commentary tool as shown
below:


Price Volume Crossover (Custom Study “PFPVC”) patterns are an attempt to capture a
pattern of crowd investment behavior prior to a significant rise or fall in prices. Marc Chaikin and I
studied 24 of these patterns and found 4 that were significant. They are located using this indicator.
The plot below shows how a price view and volume view of stocks may be combined into a “price‐
volume” view.
PFR Paper: Forecasting Stock and Commodity Prices using Price‐Volume Crossovers, Kaufman and
Chaikin, MTA Journal, 1991
Recommended Usage: Best used as to identify significant upcoming directional price movement


Output:
The red plus signs mark bearish predictions as of the close of the bar where they are plotted, while the
green plus signs mark bullish predictions.
Performance:
Excess Peak Half‐Life Sample

Name Parameters Direction
%Winners Duration Duration Size
PriceVolumeCross None Buy 5.02% 1 bars 9 bars 89079
PriceVolumeCross None Sell 5.68% 4 bars 16 bars 89079


SmartChannel (Bloomberg Study “PFSC”)
Parallax’s SmartChannel study uses an advanced geometric algorithm to quickly find and display the
most statistically significant parallel price channels. Users have the option to extend these channels into
the past or future, and also plot offset log‐periodic cycles when price breaks out of these channels.

Note the log‐
periodic offset
projection lines


This “Settings” menu allows users to
customize SmartChannel

ExtremeHurst Science
This section covers the basic science and the empirical results from our ExtremeHurst™ market predictor
model. Our research started in the late 80’s with the observation that market series appeared to have
no unique trend. An uptrend on a 5 minute chart might be embedded in a daily scale downtrend, and a
monthly scale up trend, etc. It also made intuitive sense that no unique trend exists. After all, if it did,
then all investors would quickly exploit it for profit. So following this line of thought, we wondered what
would happen next if a financial series simultaneously showed a measurably strong trend on at least two
adjacent time scales, or even three. This is a deceptively simple question, but how is trend measured?
How much trend is a strong trend? Are there preferred time scales? How much data is needed to
evaluate trend? The answers to these questions led eventually to the discovery that elements of chaos
theory, namely self‐organized criticality [13, 30] and discrete scale invariance [25‐27, 30] (discovered
independently and named by Didier Sornette), have a statistically significant predictive power in
financial markets and elsewhere. Here are the highlights of the journey.
Most scientists and engineers graduating in the 80’s or earlier were trained that experimental data
usually approximated some smooth function, but with a bit of random measurement error. To the
untrained eye, stock price data looks very similar to a smoothly trending signal covered up by a large
amount of random noise. Even the price returns appear to resemble a Gaussian distribution, which
often implies a random process. The standard engineering approach in cases like this would be to fit a
regression curve to the market series and write off the remainder to random noise [1]. Once a smooth
curve was found that fit prices well enough, one would just find the slope (“trend”) at each point. The
trend would be unique by definition, and its direction would dictate the trade direction.
Figure 10. Typical regression curve fitted to data with the remainder attributed to random sources
If this sounds too simple to be true, you have graduated to the rest of the story.

Chaos theory covers the dynamics of feedback systems [3, 8, 12, 18, 22, 30, 31]. Generally, feedback
means that the outputs of the system are fed back into the system as new inputs to produce more
outputs….and then this process continues until the loop is broken. Feedback can be negative, like in the
case of a thermostat, or positive, as seen by placing a microphone near a speaker. In the financial
markets we can imagine that price movement, news, or possibly influential people produce investment
bias “output” that is fed into the brains of the investor community, who then produce new outputs in
the form of buy and sell decisions. If this bias is positive, buying occurs, which leads to higher prices and
more positive bias, etc. Feedback systems exhibit certain characteristics, which strangely enough, seem
to match the kinds of phenomena that market technical analysts have attributed to markets over the
last 100 years. Self‐similarity is the primary example.
Financial series appear to have the same "look" on many different time scales. This is called self‐
similarity, and it happens to be one of the best ways to confirm that a series is governed by the rules of
chaos theory. Put another way, the market is what Mandelbrot [14] called a “fractal.” Not quite like a
fern or a snowflake, but more of a statistical fractal like a shoreline. We chose a fractal function
developed by Weierstrass to model a market series [8]. Weierstrass functions are continuous, meaning
you can draw one without lifting your pencil, but they have no unique slope. They look like a cycle
inside a cycle inside a cycle as shown in the next figure. Market technicians call this effect the “Elliot
Wave”, but self‐similarity has a much deeper theoretical meaning firmly based in non‐linear dynamics.

Figure 11. Weierstrass functions[8] show nested cycle patterns. It has no unique slope at any point.
Equation:
  , 1<s<2,   1
f (t )  k 1  s  2 k sin k t

According to our model, a 5 minute chart will exhibit a different slope at a given moment than the same
chart viewed on a 60 minute, daily, weekly, or monthly scale. It all depends on the length of your time

sampling. Again, this is nothing new to traders, but it represents a radical shift for scientists and
engineers. Faced with a non‐unique trend, we decided to make multiple measures of trend on different
scales. This naturally led to the research question of what might happen if they all agreed. To measure
trend we used a special local form of the Hurst exponent [35]. The Hurst exponent has a simple formula
and ranges in value between 0 and 1. Values above ½ indicate that there is a trend to events, with very
high values (towards 1) corresponding to strong trends, with a marked tendency to persist. Values
below ½ correspond to mean‐reverting behavior, which becomes more and more tightly constrained as
the exponent approaches 0. The formula for the exponent is:
Where is the range of the cumulative deviations from the mean divided by the standard
deviation, c is a constant and H is the Hurst exponent. The following picture shows the Hurst exponent
placed on a percentile rank scale of our choosing.

Figure 12. This shows the Hurst exponent percentiles and our
interpretation of the various domains.
The following picture is an example of what a time series might look like using different Hurst exponents
[34]:

Using the Hurst exponent, we are able to identify price swings which have high autocorrelation and
resemble power law upswings or downswings such as the one shown in the following picture [30]:
There was another key timing factor in addition to the parabolic price move which we literally stumbled
upon while trying to limit our Hurst sampling look back periods. Cycles were present during these
parabolic moves, and by exploiting them, we were better able to time the critical point. The next
section discusses log periodic cycles and the science that ties it all together.
Extremes of investor behavior are most evident in large scale market manias or crashes, but like
earthquakes, significantly more low magnitude events occur than large ones. It is our contention that
markets set themselves up in critical states through feedback, like earthquakes, and then we see
tremor‐like activity of all sizes and scales at both highs and lows. This area of scientific research is called
“Self‐Organized Criticality” or SOC for short. Bak, Tang, and Weisenfield (BTW) [36] first introduced SOC
in 1987 by studying the occurrences of avalanches in sand pile models. It was their contention that large
systems with many constituent parts, organize themselves into states that resemble systems in
equilibrium at critical points.
In the case of the markets, investors are influenced by news, price changes, and other investors within
their local social network. This influence conditions their willingness to buy or sell, which in turn affects
price and influences others, and so on. One of the methods used to understand the dynamics of large
critical systems is called the Renormalization Group Method [10, 13, 27, 30] this method uses scaling to
make sense of complex systems. We have applied this method to learn some basic principles about

markets: First, it takes very few net buyers or sellers to trigger a buying or selling trend, or end one.
Second, the buildup to a critical point resembles the power law function C1*(1‐ t/tc)x , where C1 is a
constant, x is a negative power, t is time, and tc is the time of the critical point. Thirdly, there is often a
log‐periodic ripple in price that converges to the critical point that has the form: 1+C2*cos[ p*ln(1‐
t/tc)+q ] , where ln(x) is the natural log function, cos(x) is the cosine function, and C2, p, and q are real
constants[30].
In the time domain the small ripple signal we search for looks like this:
While in the log time domain it becomes a sine wave:
Sornette [26] attributes this ripple to what he has called “discrete scale invariance” which means that
the price series is only scale invariant at certain discrete times. He attributes this effect to fractal
dimensions being complex numbers instead of real numbers. Translation: Discrete scale invariance
occurs when all trend measurements at all scales agree with each other. Our filter is designed to find
these critical points.
The filter we have created finds these critical points at the end of power law advances and declines.
These we call “extensions.” There is another critical point not yet written about in the literature that we
call “compressions.” A compression has the same log periodic behavior shown above, but without the
power law advance or decline. A compression occurs at points of extreme mean reversion, and is
followed by the start of a new trend, usually with a volatile breakout.

ExtremeHurst Testing and Statistics
ExtremeHurst signals predict market critical points which mark a change in trend, or in mathematical
terms, increasing price curvature. There are three types of ExtremeHurst signals: Top and Bottom
Extensions and Compressions. The predicted price behavior following a top extension is for price to
cease moving up. Likewise, for bottom extensions we expect price to cease moving down.
Compressions are preceded by flat periods, so the post‐signal period must see price move away from
current price and begin to trend:

These charts show the simulated periods before and after each of our signal types. The green lines
show no trend change across the signal. We have created a tool to measure curvature called the Turn
Measurement Index (TMI) which is an approximation to the second derivative of a series using the
following differential formula from Calculus:

We have manipulated the formula to work with financial data on all scales using a “random walk”
assumption. The final form looks like this (Taken from a Parallax research document):


The next picture is an example of the TMI formula applied to Microsoft stock prices over lengths of 7

and 15 days:
Note that the 15 day TMI picked up the large trend change, but the 7 day TMI did not see it.

We expect that there will be significant net increase in TMI across all of our signal dates when compared
with the set of all potential signal dates. We have applied the Students t‐statistical test to measure this
hypothesis. First though, we will introduce the weighted average of different length TMI measures,
called TMIw. We need multiple measurement lengths since using only one length is likely to miss some
trend changes. We chose lengths ranging from about 1 to 6 weeks, with an average of 13 days. This
length is close to the length over which ExtremeHurst signals are known to have an effect. In the next
graph we have calculated the daily TMIw for a random set of S&P 500 stocks over 20 years and made a
histogram. Remember that high TMIw means that there was a very high amount of trend change across
a given date.
We have a slightly different need however. We need to know how much net trend change occurred
across a given date. This is simply calculated taking the difference between the two‐sided (pre and post‐
signal) TMIw above, and a one‐sided TMIw which assumes flat post‐signal prices as shown below.

When we plot the histogram of the net trend change for potential Extension days, and then compare it
to the net trend changes from our actual Extension signal days, we get the following picture:

In the plot above, the amount of net trend change increases as we move to the right, so our signals,
shown in blue, clearly have more “bend” than the potential signal days. The potential signal days are
those days whose one‐sided TMIw is greater than or equal to the minimum of that seen on our actual
signal dates. In other words, potential signal date have strong trends leading into them…..but the
market fails to bend as much coming out of them as it does when an actual Extension signal hits. Our
statistic measured p<=5.63x10‐6 which means there is less than a 1 in 177,000 chance that the two
histograms above were generated by the same process. In the few pages which follow, some additional
tests are shown for a prior version of ExtremeHurst.

ExtremeHurst Testing and Statistics (older)


The question usually arises as to how ExtremeHurst signals effect standard indicators like ADX and
Implied Volatility. ADX is a directionality measure, so extensions should lead to a drop in ADX, while
compressions precede directional movement and so should lead to higher ADX values. We have
confirmed this.

The next plot shows average implied volatility (from options) before and after the three signal types.
The red line is due to extension bottoms, and shows that volatility drops 32% following a bottom signal.
The green line is the average volatility before and after a top extension. In this case we see a 10%
increase after a top. There is no change at all in volatility around a compression signal, and volatility is
also relatively low.


ExtremeHurst Users
Several money managers are using or have used the ExtremeHurst product as a part of their
management business and have consented to be listed as references for potential users:
1. MARTIN MCBROOM, PRESIDENT
MCBROOM & ASSOCIATES, LLC.
6795 EDMOND ST. #10
LAS VEGAS, NEVADA 89118
(213) 612‐4626
EMAIL: [email protected]

2. BILL MECKEL, PRESIDENT
MARQUE MILLENNIUM CAPITAL MANAGEMENT
850 THIRD AVE, FLOOR 18
NEW YORK, NY 10022
(212) 759‐6800

3. MARK ASTLEY, CEO
MILLENNIUM GLOBAL INVESTMENTS LTD.
CASSINI HOUSE
57‐59 ST JAMES STREET
LONDON SW1A 1LD, UNITED KINGDOM
+44(0) 20 7663 8910

4. RAJPAL ARULPRAGASAM, PRESIDENT
ARCHETYPE RISK ADVISORS, INC.
195 CHURCH STREET, FLOOR 9
NEW HAVEN, CT 06510‐2009
(203) 497‐8700


ExtremeHurst For Bloomberg Guide

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ExtremeHurst For Bloomberg Guide

Uploaded by

Copyright:

Available Formats

Parallax Financial Research

Figure 1. This is an example of a weekly scale Extension top signal.

Figure 3. Time series generated with different Hurst exponents

Figure 5. Extreme investor behavior.

Figure 8. A harmonic oscillator is a great way to envision a compression and extension

Figure 9. Examples of each ExtremeHurst signal type

Marker Symbol Float over

Recommended Usage: Best used as to identify significant upcoming directional price movement

Excess Peak Half‐Life Sample

The next picture is an example of the TMI formula applied to Microsoft stock prices over lengths of 7

You might also like