NASDAQ Stock Market Timing

How can I trust that the historic performance record is genuine?

This should be the first question asked by anyone who visits this website. The website's goal is to demonstrate our real-time performance and we are not looking to sell you use of the model.

The model was initially developed in 2001 but it was in late 2002 that we began sending out real-time model readings to contacts in the hedge fund, trading and financial sector.

At that time the model was in its early stages of development and although our signals were accurate, it could not be proven whether this would translate into real-world trading success.

So after nearly two years of broadly accurate signals, we had a conversation in 2004 with a senior executive in the hedge fund industry.

He suggested tracking real-time performance using a notional $1m of capital and a rigid framework of rules.

The rules of this performance tracking are shown in each historic report produced since May 2004, which you can find using the yearly performance tabs on the left (look under the “Trading Model Rules” section in any past report).

In short, we must write a report every weekend (even if no trade) and provide the latest model reading to our referees. These include Jack Schwager, a renowned expert in assessing trading models.

This is always done in advance by setting market (i.e. at the next market open), limit (at a predefined level) or stop orders.

The use of market, limit or stop orders means that all trades could really have been placed. If stop or limit levels are missed due to a gap up or gap down, then we are filled at the "real level" not the theoretical one.

Indeed, our trades are subject to missed fills, narrowly missed stops, gaps up/down and any number of factors which any seasoned NASDAQ trader will recognise.

We are meticulous the following week in documenting the level at which the actual trade took place, even if it is different to the order level set (for example if the market gaps up or down beyond the expected trading level).

This system of rules helps to safeguard integrity and may even have constrained the model’s performance, because we are not able to trade during the week nor achieve optimal prices if a signal is generated midweek (we must wait for the following weekend to place the next week's order).

We have nevertheless stuck to the rules rigidly and would be very willing to have this performance audited. Every documented trade may be checked against the real market levels at the time right down to the day of execution by tracing the individual real-time reports.

All our referees (located in different parts of the world and receiving the report simultaneously by e-mail) can vouch for the fact that trades are always signalled in advance.

We've never missed sending a weekly report out to our referees since real-time performance tracking began in May 2004.

The date on which each report was emailed out is also documented in the footer of each report.

Has there been any back-testing done at all?

We’ve deliberately avoided back-testing the model due to the problems of curve fitting and hindsight bias. Even at a relatively early stage (monthly reports we did predating the weekly ones and covering the period from October 2002 to May 2004) we sought to improve the model's robustness only through "live trading".

This is precisely why today we have such strong personal conviction in the model's ability to withstand a wide range of market scenarios.

To reiterate, all performance is generated through actual live trades.

What is the intellectual basis of the model?

ContraQuant is based on age-old, time tested principles of contrarian sentiment as espoused by Humphrey Neill (in his timeless classic, “The Art of Contrary Thinking”) and developed by others.

The basis of the contrarian approach is that bearish turning points occur when the consensus is most bullish . When market participants are asked about their market expectations they will be most bullish after they have bought heavily.

This is obviously a natural human reaction, since each of us feels compelled to rationalise and justify actions we may have taken.

However, when this feeling is a collective one, and everyone agrees that the market will rise, this is often likely a bearish turning point and it is probable that buying power has been (at least temporarily) exhausted.

The key difficulty, of course, is in picking up such turning points with precision. For example, how does a contrarian investor avoid getting cut by the so-called "falling knife" if he/she wants to go LONG?

That is where the ContraQuant's statistical timing engine comes into play: it absorbs a wide range of specially-chosen sentiment data, assigns a weight to each input factor and comes up with a summary probability of the market's next directional move.

We do not currently attempt to forecast the pricing of individual stocks, although there is sentiment data available to attempt this.

Isn’t contrarian investing just common sense and widely practised already?

Many fund managers aspire to be contrarians, and it is often said, for example, that the best time to go long of the market is when there is “blood on the streets”.

In practice, however, both private investors and professional fund managers often get their timing wrong. Many profess not to believe in market timing, despite the evidence that markets move cyclically.

So is it any wonder then that an estimated 80% of professional fund managers underperform their benchmark index? And they sometimes charge steep management fees for the privilege.

Being contrarian is easy to talk about but hard to implement. Do read through any of our more than 240 historic weekly reports for a flavour of how we put our own brand of contrarianism into practice.

It is clear from those past reports that our trades generally went against the market consensus of the time.

For an example, think of our trade entered in early 2008. Admittedly we entered this LONG a few weeks early (when our model picked up around an 80% probability of a rise).

However, it is clear that most investment commentators at the time seemed very nervous of the market (precipitated by the mortgage crisis, the high oil price and other factors).

Indeed, they were expecting almost to be ridiculed for being LONG, a time when the model indicated that it was in fact most rational - and least risky - to take such a stance.

We closed this trade successfully on 20th May capturing a double-digit percentage profit (click 2008 Performance and see the reports for further detail on how this trade was spotted, entered and closed).

How the early 2008 trade was opened:

http://www.market-timing-signals.com/SQ Bulletin 18Jan08.pdf

http://www.market-timing-signals.com/SQ Bulletin 25Jan08.pdf

http://www.market-timing-signals.com/SQ Bulletin 1Feb08.pdf

http://www.market-timing-signals.com/SQ Bulletin 8Feb08.pdf

How the early 2008 trade was profitably closed:

http://www.market-timing-signals.com/SQ Bulletin 16May08.pdf

http://www.market-timing-signals.com/SQ Bulletin 23May08.pdf

Of course, it's also easy to be contrarian and wrong. That is where the statistical science behind the ContraQuant model is the key to its relative accuracy in picking up the timing of market turning points.

It is not just the fact that ContraQuant is a contrarian model, it is also its robustness in quantifying its own predictive accuracy which makes it powerful.

The model actively tells you from the outset how much confidence it has for a trade, captured in its so-called "bullish probability".

Even with the consistency of the track record so far, we would never claim to get it exactly right as the model never provides anything more than a probability. Indeed, the loss incurred in our late 2008 trade - the first ever - confirms this fact.

How does one interpret the summary reading of "SQ"?

“SQ” is a market timing indicator developed to quantify the probability of a bullish stock market move over the short to medium (intermediate) term.

It is based on a unique approach through the adaptation of an established marketing research technique, to derive a summary index of market sentiment.

SQ is then calculated to give a percent probability of the market going up or down in a 2 week to 4 month time horizon.

The model gives buy signals when the market consensus is overtly bearish and sell signals when the consensus is predominantly bullish.

It predicts the intermediate trend of the stock market with great accuracy by highlighting bullish and bearish “capitulation” points.

Moreover, it gives a quantifiable probability of these turning points which allows for "scaling in" and money and risk management.

Read through any of the hundreds of reports filed on this website to get a flavour of how this actually works in practice. It has been profitably implemented on 10 separate occasions in real time since May 2004, with only 1 incorrect signal given.

How is the model indicator derived and what does it predict?

SQ is a composite measure derived from several sub-indicators of market sentiment identified and tested over more than 7 years of research and development.

Some indicators relate to the options market whilst others combine “intention” and “behaviour” by market participants. Still others are proprietary measures we have created ourselves.

All model inputs are tried and tested guages of market sentiment. SQ is a forward-looking index optimised to predict movements in the Nasdaq 100 and its close twin the NASDAQ Composite.

These specially-chosen inputs drive the "engine" of our model and we track them constantly.

They include various time-tested measures of market sentiment including, but not limited to, the following: market surveys, asset allocation measures, volatility measures, options market indicators and so on.

Several inputs have been combined and melded together to create our own proprietary indicators.

All inputs are weighted and combined into a single, transparent index called “SQ” which measures the bullish probability of the market over a 2 week to 4 month future period. This SQ index usually moves inversely to the market, though crucially not always.

Indeed, we have benefited several times from going LONG on a pullback in a rising market as our model shows many investors to be “worried” by it rising “too fast”. Contrarians know this to be a classic sign that there is money waiting to enter on the sidelines.

As the market climbs such a wall of worry we may remain safely LONG. Once this slips into excessive bullishness (or even shows early signs that it might) we close out. If this then in turn becomes euphoria we may potentially reverse and go SHORT.

While the sentiment inputs are themselves part of the intellectual capital of the model and several of them are fairly obscure, there is nothing particularly surprising about them.

Other market forecasters probably use some of them (though not our self-created proprietary inputs).

The core technology of the model is actually in our continuous R&D in examining, optimising, rescaling and reweighting indicators while retaining the integrity of the model's basic predictive accuracy.

That is why we are not afraid of anyone trying to copy our system, because it is simply impossible to replicate this R&D occurring over the last 7 or so years.

How are profits made?

Profits are made because markets are fundamentally driven by emotion. Our proprietary composite SQ measure quantifies the predominant emotion of market participants on a scale of 0 to 100.

When everyone is bullish and buying power is exhausted we go short and vice versa. There is relatively little risk that our model will be correlated to the underlying market trend.

There are however certainly times when the market trend is bullish and we’ll be LONG because market sentiment indicates that the bullish trend is being met with extreme scepticism. It’s the model’s job to measure this.

This is an important point, because it confirms that the model is not blindly contrarian but rather picks up on subtle changes in sentiment to decide which way to be positioned.

We will generally enter a trade a little "too early" and may also exit "a little early". The model aims to pick up the sweet spot of the market's directional move, rather than greedily trying to capture every point.

According to the model, the position is most secure when it is fairly lonely and contrary to the perceived wisdom and expected market direction.

When our positioning becomes more crowded and sentiment is mixed (SQ between 45 to 55%) we would tend to exit any trade and flatten our position.

Protective "emergency stops" are used for risk management at all times, to legislate for totally unforeseen "out of the blue" exogenous events.

Remember that it is the objective statistical basis of the model that makes the signals solid. We are not blind contrarians but rather patient and often reluctant ones, only convinced by hard data.

Why are issued signals and trades relatively infrequent?

This is a function of the model's construction and natural make-up as already described. Most of the model's sentiment inputs carry a slightly longer time horizon than merely hours or days.

The model's job is to identify only the highest-probability trades and this can be done only when there is a genuine intermediate-term "extreme" of sentiment.

We have found that this typically happens 1 to 3 times a year, though it could occasionally be more (or less) frequent.

As the Track Record confirms, this infrequent high-probability trading strategy has been profitable, though not entirely immune to downturns as the late 2008 experience shows.

We see no reason currently to issue more frequent signals, particularly as the model has generated above-average annual returns through just one or two trades per year. Do inspect the Track Record for further information.

It is certainly not advisable to use the model's signals to trade over a shorter timeframe, though it is plausible that there may be scope for further research to construct options and other trading scenarios around this.

What is the model’s edge?

Our edge is in the model's inputs and the way they are combined and weighted. We use well-established inputs that have a track record of decades, not all of which are widely-known.

However, it’s the way that inputs are combined that is the true source of the model’s predictive power.

Some inputs work for a certain period of time but it is not a tenable strategy to follow them in isolation. This applies particularly if they become closely watched by the consensus of market participants, at which point they begin to lose their contrary nature.

Some other timing models track far too many inputs without fully evaluating how effective they are. Using a very large set of inputs can also give conflicting or confusing signals. We on the other hand take a rigorously selective approach of "less is more".

Hence our creation and selection of both proprietary and tailored input measures (allocated a weighting proven to work over a significant period of time) is a very important edge of the model.

Another edge is the model's transparency and the fact that investors could potentially apply a variety of strategies with it. With the ContraQuant model, investors always know where they stand and can quantify the riskiness of any given trade in advance.

A further model feature is the transparency of the underlying instrument traded - the NASDAQ - which also happens to be one of the most liquid trading vehicles around.

Compare the ease of entry and exit to a NASDAQ index trade with having to manage a position in an illiquid stock.

The liquidity and innate capacity of the NASDAQ also make the ContraQuant model a highly scalable proposition.

Over the last few years the model has picked up several of the market’s major directional moves with precise timing. Just browse the historic performance tabs to read the evidence.

This proven predictive capability could certainly be even more deeply and profitably exploited.

Has the model performance experienced significant drawdowns?

The model experienced its biggest ever drawdown in September/October 2008 when it was hit by the financial storm which impacted all the world's markets. An incorrect LONG signal was given which led to a 20% loss.

This highlighted a key dilemma of the model, the fact that it is very hard to pick the exact turning point in a market displaying extreme sentiment. Especially because the model gives an even stronger signal at the time the stop is hit.

However, if one looks at the longer-term historic performance chart at the top of the Track Record page, it is clear that periods of drawdown have been relatively infrequent.

An important feature of the model is that we can predict the times to expect drawdown, namely just after we have entered a fresh trade.

Because our model and its signals are contrarian we are always going against the short-term trend and actually expect there to be a period of drawdown initially.

This predictability of drawdown is useful from a risk management strategy as various strategies could be developed to counteract it. In practice, however, the ContraQuant model's predictive capability has been precise and we have experienced only rare drawdowns.

Typically our drawdowns only last about a month (predictably just after we enter a trade) and then are quickly recouped straight after. The exception to this occurred in late 2008.

Apart from this, the only other significant drawdowns are described below.

In June and July 2006 we experienced a drawdown of 4.9% over 2 months but quickly recouped this in the subsequent period (ending up with a 10.5% profit on the trade).

In February and March 2008 we experienced a total 2.5% drawdown but then in April alone we gained 7.2% and overall ended up with 12.2% profit on the trade.

So in terms of average duration, frequency and depth of drawdown the ContraQuant model generally outperforms most other strategies.

How else might the model's signals be used?

We believe that one of the prime strengths of the model is its ability to manage risk and avoid getting involved in the last throes of an intermediate trend.

Therefore the ContraQuant signals might be of value to those who trade the NASDAQ index (or indeed stocks) from a LONG-only perspective, in that it can assist in timing entry and exit from individual trades.

There is well-established research which shows that the bulk of a stock's short and intermediate term price movement is accounted for by the trend in the market as opposed to stock-specific factors.

Therefore the ContraQuant signals might assist professional LONG-only managers to stay consistently on the right side of the market, especially during those key turning points which occur once or several times every year.

Or for a multi-asset trading approach or fund, the ContraQuant model could be used to indicate weighting and switching over time between different asset classes.

This is particularly valuable when one considers the infrequency and accuracy of the ContraQuant market timing signals. There is great scope for the model to help traders avoid potentially costly mistakes.

Can the model be used to forecast markets and trade instruments other than the NASDAQ?

Different stock markets around the world have different correlations with the US indices in general and the NASDAQ in particular. Such correlations are usually not stable over time.

Since our model inputs are derived principally from the NASDAQ and US markets, we cannot confidently forecast other stock market indices with any level of precision.

It may be possible in the long term to develop further sub-models based on the ContraQuant approach, once more and better sentiment input data becomes available. We are certainly open to ideas in this direction.

Is there trading discretion in running the ContraQuant model?

There is very limited discretion in generating signals, insofar as the “scaling” and “weighting” of inputs is constantly under review. In practice, however, structural model changes have been infrequent due to its long run of predictive success.

Since we currently execute trades through our weekly report, we often execute them with market orders at the next open (there’s no choice as to the precise level achieved) or by means of limit and stop orders.

When the model signals that we should close a trade, we usually insert a trailing stop to ride with any momentum that often lies behind our open position.

Occasionally, we’ll also set a limit order to benefit from any overshoot on our open position while trailing the stop at the same time.

Numerous live examples of how this actually happened in real-time practice can be seen by examining our historic trades and reports.

Click on the annual performance tabs on the top left and you'll find ample evidence to back up the model's claimed ability to exit as well as enter trades optimally.

Again, the levels and tightness of any stops are dependent on the percentage reading of the model at any given time.

However, the model is essentially mechanical in prescribing trades and we would generally not seek to override its core posture. Any limited discretion lies in our own money and risk management and maybe in our intangible "feel" for the underlying sentiment data.

How does money management of the model operate?

There is no meaningful discretion in choosing when to trade and this is by design. Once the model passes certain specified trigger points, trades must be executed.

However, there is the option of managing risk and money as dictated by money management constraints or goals, though even this operates within fairly strictly predefined parameters.

The standard approach has been to “scale in” and “scale out” of trades according to the probabilistic scale dictated by the model.

In theory this would imply 100% exposure only when the model reaches a theoretical reading of 100%...in practice however we have scaled in our full capital at levels below this extreme (around 80% or sometimes less to go LONG and perhaps at SQ=20% to go short).

The aggressiveness (or not) of entry is a function of the conviction behind the trade, captured not only by the headline SQ reading but also by our years of interpreting the underlying sentiment data.

How is risk managed?

Risk management is automated. We put an “emergency stop” in place whenever a new trade is opened, however this naturally needs to be some distance away because the model will almost always signal against the immediate short-term trend.

Indeed, we always "expect" the trade to go against our initial position to start with since we are generally swimming against the short-term tide.

In practice, however, careful examination of the model's historical live trades since 2004 reveals that the model has often picked significant market turns within a few days or points.

We would be the first to admit that this has not always been by design. Nevertheless, the overall accuracy of the model in picking its entry and exit points cannot be disputed.

In theory the model at the stop level would indicate an even higher probability trade, therefore we believe it right to keep a considerable distance. In practice, our “emergency stops” have only been hit once, which was in late 2008.

These extreme stops are usually significantly tightened and become "trailing stops" as our model turns neutral (typically readings of 45% to 55% bullish probability) and we begin to prepare ourselves to close the trade.

We are constantly researching new ways to improve the model's risk management mechanisms and welcome further thoughts and ideas in this direction.

Do also read the Articles to learn of avenues we have already been exploring.

Can you provide an example of a successful trade?

Market sentiment can shift very quickly and for this reason we always keep a close watch on our model inputs, whether or not we are currently in a trade.

For example the sudden drop in the market from late February 2007 was accompanied by a surge in bearish sentiment (as reflected in our model’s reading shooting up above 80% by mid-March).

Therefore while many market participants were too nervous to call the direction of the market, we entered a confident LONG position close to the lows for the year (at a NASDAQ Composite level of 2345).

The following 2 real-time reports show how we opened this trade:

http://www.market-timing-signals.com/SQ Bulletin 2Mar07.pdf

http://www.market-timing-signals.com/SQ Bulletin 9Mar07.pdf

We closed that same trade less than 2 months later after capturing a full 200 NASDAQ points of profit. Here are the 2 reports setting up and documenting closure of this trade respectively:

http://www.market-timing-signals.com/SQ Bulletin 27Apr07.pdf

http://www.market-timing-signals.com/SQ Bulletin 4May07.pdf

This is a textbook trade which highlights the model's predictive accuracy.

It is not an isolated case, as you can see if you browse through our historic reports by clicking the yearly performance tabs on the left.

Indeed we entered an equally profitable LONG trade in late January and early Feb 2008, described earlier in these FAQs.

In the past we have entered SHORT as well as LONG trades, another factor which helps the ContraQuant model's performance stay uncorrelated to overall market direction.

Given the consistent performance to date, why has a hedge fund or managed account not yet been seeded from the model?

There are probably a variety of reasons to explain this. Some of these are genuine improvement opportunities for further development of the model.

First, we have not actively sought to market the model until recently. The initial intention was simply to continuously enhance the model's robustness through real-time trading. As elaborated on below, we are in this project for the long haul.

We are not interested in only making a quick buck from the model. We instead want to continually refine the model so that it remains resilient for a long time to come.

Only now, after more than 4 years of robust real-time performance (on top of another 2 before that in initial development) do we feel ready to bring the model to market. Please scroll down to see some options under consideration.

Our confidence is reinforced by the model's solid record to date which has beaten most mainstream hedge fund strategies for quite a while now delivering performance across several types of market backdrop.

Interestingly, the ContraQuant model seems to come into its own the more that most mainstream hedge fund strategies appear to struggle, although we were not immune to the downturn in 2008.

On the other hand, August 2007 was a tough month for many hedge funds but highly positive for the ContraQuant strategy.

In this sense, we believe that the ContraQuant model is potentially a valuable diversification tool for any investor seeking consistent absolute returns, whatever the market backdrop.

This leads to a second reason, which is that our model doesn’t fit easily inside any one of the ready-made “umbrella strategies” that are tracked by the main hedge indices. This makes it hard to categorise.

On the other hand, the basic model approach is so transparent and easy to follow (everyone knows the latest NASDAQ level) that one would expect this to counterbalance this risk. Especially as many hedge funds are so opaque.

One would also think that the liquidity of the NASDAQ as a trading instrument would be attractive, as well as the mind-boggling potential to scale up the ContraQuant strategy (unlike the more esoteric hedge fund strategies which often have severe constraints).

Third, the model is infrequently traded. Therefore we may spend months doing nothing and waiting for a signal, which will only be flagged in an “extreme” environment. This may put off some investors who like to see constant activity.

On the other hand, one could argue, surely it’s the return that counts. In a market timing model such as ours, "cash" is a very valid signal and a volatility stabiliser.

Isn't it more efficient to sit and wait for the right, high-probability opportunities than put cash to work just to look active or justify high management fees?

Fourth we do not have a fancy set-up around the model in terms of an expensive back-office operation and organisation.

This is a box which many providers of capital need to have ticked but which we believe adds unnecessary cost to our strategy (cost which would again typically be passed on to investors).

Finally, the model is yet to fully resolve the standard dilemma of any contrarian model, that of risk management. The dilemma, along with solutions we have so far explored, is illustrated within our Articles.

If you have any questions not covered by those set out above then please do not hesitate to click through to the Contact page.

Can the model work in any type of market?

So far the ContraQuant model has been running in real time since May 2004 and in its earlier form, since October 2002. It has generated only one losing signal in this time.

The model sends a signal relatively infrequently and therefore only high-probability trade alerts are issued.

There have been 9 LONG and 2 SHORT trades (see Track Record for a full listing of all trades and their executed NASDAQ levels). Only one has not been profitable.

We are confident of the model's effectiveness in all environments because it changes chameleon-like to its backdrop.

Thus at times it can be momentum-following (e.g. buying dips in a bull market) and at other times it will signal a long-term turning point on either side. A range of scenarios has already been captured if you examine the historic real-time track record closely.

The model’s success is therefore not tied to any one type of market. It can sometimes follow a trend and at other times fade against one.

If you trace the live trading record you’ll see that LONG and SHORT trades have been successful, both in range-bound and trending markets.

Sometimes we have scaled in quickly on a sharp dip/rise and at other times we have gradually accumulated.

The model depends entirely on market sentiment, specifically on pinning down extremes which signal turning points.

Often significant turns in sentiment can occur even in a trendless market and the model has proven itself sensitive enough to pick these up.

Sentiment is measured against its context: for example, the threshold for “bullish exuberance” is naturally higher in a long-established bull market than it would be in a bear market.

Again the model is sensitive enough to pick up these subtle changes in context.

Of course, we would never claim to identify all the turning points in a market.

We only aim to pick up quite a few of them (on average about 2 every year) to generate a consistent double-digit annual return which is uncorrelated to any other strategy.

What is the long-term vision for use of the ContraQuant model?

Our model's performance has generally been ahead of most hedge funds currently in the market.

It has also retained the important quality of being fairly uncorrelated to most mainstream and niche strategies.

We believe that performance ultimately speaks for itself and have no doubt that the model will be seeded into some kind of fund or licensed in some manner (see below).

We are nevertheless keen to be selective about any opportunity ultimately taken in commercially exploiting the model.

Our vision is that the model should never be subject to investment fads and should always retain its sharp predictive capability, by sticking to the time-tested fundamental principles of contrarian investing.

We are therefore laying the foundation for the model to be successfully developed over years and hopefully decades. We are willing to wait for the right opportunity and partner to work with us.

If you or your organisation shares this long-term vision then do Contact us.