Diversified Trend Trading Approach

Just over the boarder in NSW, special bit of the country

Hence your avatar? Pretty spec countryside. :-)

How to Detect Real Trends from the Noise - Thought for Sunday

Previous posts have discussed the difficulty in interpreting trends arising from simply a random string of higher highs or lower lows versus those trends which display true directional tendency (auto-correlation or levy drift) over time. Obviously as trend followers we will be trading both forms of trend as we simply follow price, however it is only the latter (if it exists) that will lead to alpha. So how do we distinguish between the two in order to filter out the first and ride the latter?

Visually both form of trend are very difficult to tell apart. The only visual feature that may assist in discriminating between the two is the directional extent of the price action which when your are trading from the right edge of a chart, is very difficult to discern. Of course it's easy to determine trends of significant directional extent 'after the event', but how do you know the time to jump into a trend and take a ride? Trend followers are always late to arrive to a trend, but wouldn't it be nice to have a method to let us know with more confidence when to start hitching a ride.

One feature I have noted that occurs with regularity (say a few times each year on average) in my discretionary system portfolio is that occasionally markets become more correlated and it is during these times that I notice that trends are far more widespread than normal in my portfolio. When something significant is happening in the market, then most of the instruments in the portfolio start moving uni-directionally (whether up or down). Another noticeable aspect of performance with the discretionary portfolio is that most of the times I take a hit from an incorrect decision, is when a single instrument appears to trend without 'trend' confirmation from the balance of instruments in the portfolio.

What does correlation mean? Well it is one sure fire way of telling me when there is something 'real' going on in the markets as opposed to random jitters. The inter-dependencies that exist across assets and asset classes is a symptom that something fundamental which triggers coordinated participant behaviour is going on, such as a flight to liquidity. If you are a betting man/woman, this is when you would have a more likely chance of hitting the jackpot as there is a one-sided bias to participant behaviour.

These conclusions are telling me something. Correlation perhaps is a trend followers friend....

Maybe we can employ a filter across the entire portfolio itself that measures portfolio 'trendiness' and restrict our trade entries to when those 'window's of opportunity' are opened by the filter.

Typical volatility measures are useful but can be misleading in that what we really want is a filter that discerns uni-directional price movement across the portfolio as opposed to mere random volatility (that could either be advantageous or disadvantageous to the trend follower dependent on the nature of the volatility).

Perry Kaufman's signal to noise ratio comes to mind as an indicator (KAMA) that perhaps could be enhanced and deployed for these purposes across a portfolio. Here is a link to a useful indicator of Perry's that with a few tweaks might have portfolio application for the diversified trend followers out there.

Alternatively a measure of overall portfolio correlation might be the way to go in that as the modulus (absolute value) of total portfolio correlation increase towards 1, then the green lights go on for trade opportunities.

Anyway, something to munch on......

Regards

C

Bwahahahahahahahah

The era of stupidity greets us again. Put the crash helmets on, this cycle is going to be a corker (GFC^2). :-)

Credit fuelled bubbles....tell em' they're dreamin'. We're as safe as houses....*literally*.

nicee

"We only pick stocks from the S&P 500 index. Historical membership accounted for of course." survivalship bias





found that on the terminal

"If only we'd stop trying to be happy we'd have a pretty good time."
Edith Wharton

Clenow has a large axe to grind with the mutual fraternity. I don't blame him. The SPIVA scorecard for these guys is appalling. I just don't understand the stupidity of investors. The spin from the classic buy and hold manager supported by the CNBC's of this world etc. is toxic.

A related article from Clenow.

I know few intellegent guys who made a fortune in garbage business. They let their moneys get invested by this mutual fund skunks. What I noticed is that they get them by just telling this few starters:
You saw what China did? ...
You saw what Fed did? ...
Have you seen what Saudis did? ...
Have you read about ? ...

Of course they dont know, even if they do...they all like to talk about stuff, and be the part of the financial club. They have a few skunks which trade their moneys institutionally. Know this guys, they like to buy the dip kinda guys

but the worst are the guys: we will have inflation in 20%s you need to buy yourself gold....they will be gold bulls at any number at any DD. Those are the dangerous ones..

but there are worse...new breed...cryptocurrency guys...you need to hear and see their investment progression curve

eeeeek. Run for the hills I tell's ya.

Raised from the loins of Hydra and blended with scorpion venom....I present to all the prudent investors out there ........the 'Bitcoin broker'

...He took our game for his own need.

From what I could find on the Internet (so I may be plain wrong) LTCM didn't fail because their system was bad. They bankrupt due to human mistakes.

Their method was comparable to an arbitrage. Arbitrage work when you are discreet. They were trading by the billions! They became so big that they became the market. They were the inefficiency. They became "traded". Human mistake.

Their model predicted that what happened couldn't happen in a 10 millions year period. Yet it did. When you are a Nobel prize and you have got a piece of evidence that something that can't happen according to your model is just happening you have to recognize that the model is proved out of its boundaries. They should have stopped their trading. They continued without even reducing their exposure. Human mistake.

Should they have stopped trading for sometime and resumed later, after the dust settled, they may have discovered that the arbitrage was gone. But if their unique 30% p.a. system was no longer working how can they have differentiated from their competitors? How to explain your share holders that you'll now just do 7% p.a. like the others?

I think greed+ego+politics was a much stronger reason for their failure than their mathematical model.

Good points Pip

I was reading "When Genius Failed" by Roger Lowenstein and noted how similar the story was to the credit default swap and CDO debacle that was popularised in the 'Big Short". At least their was a bit of opposition from Scholes that led to the debacle. They abandoned their traditional hedging strategy (as stated for a portion of their portfolio) which was a signature staple at LTCM and also their bond spread arbitrage strategy as liquidity dried up with their massive growth and went full blown into their very high leveraged equity vol strategy using options. As you say human error was obviously rampant. The point was however that the style of strategy used was massively exposed and open-ended with respect to risk and despite the brightest of minds, they fell for the age old trap of FIGJAM. :-)

The following excerpts from the book " When Genius failed" pretty well sums things up.

Discussion around Backtesting Trading Platforms and Statistical Arbitrage Methods and More

An interview by Andrew Swanscott from Better Systems Trader and Nitesh Khandelwal

Some key statements drawn from he transcript:


















Enjoy

C

Full transcript attached.

Following on from recent posts discussing Statistical Arbitrage, here is a good article that reflects the subtle but important differences between co-integration versus correlation which is essential in understanding with respect to pairs trading (stat arb techniques) where the strategy is dependent upon reversion to a historical mean. Remember however that in theory, the mean is stationary whereas in reality, a non-stationary mean is common-place adding what I refer to as 'reality risk' to these models.

While beta and delta risk are effectively eliminated through stat arbitrage methods (which is a delight to observe in action in volatile markets), it is not risk free. A change in market conditions (non-stationery risk), one sided risk (industry risk) associated solely with a specific instrument in a pair (such as de-listing, a specific industry related risk etc.) , execution risk (risks associated with order execution latency), model risk( eg. with HFT, many models utilising the same strategy will harvest any available alpha from the method) need to be closely considered.

Being a negative skew strategy, success is dependent on increasing the size of positions as the relative price differential between the pair takes an excursion from the mean, which leaves the trader particularly exposed if any of the risks above unfold....hence the pressure cooker analogy of post 879.

Regards

C

A Bit of relief to the Daily Grind

It is tempting to follow most of the hedge funds into negative skew strategies due to high sharpe ratios delivered as the downside risks are frequently forgotten until it is too late.

Here is an interesting interview with Nigol Koulajian from Quest Partners LLC.

Performance results since 2002 are as follows:



Cheers

C

hi @copernicus ,ı think about your adding more with when price go further and start with min size about week or so.and ı decided to apply it to my demo trading and test it.ı in fact do the oposite.while the price go further my favor ı close partially.this will be very big chamge and it will take time to adapt ı am sure but this starting small and adding more sounds good after some think on it.at the end ı am sure ı will come to a point where ı satrt with small and add more and at some point start closing and exit all.some kind of bell curve ı think.tdata will tell.greenpips

Let me know how it goes V. It seems to bear fruit with the discretionary approach I use in this thread as it is a different method than traditional pyramiding into positions which I am not a big fan of. The benefit of scaling into positions using different time-frames is that you are ensuring that momentum is in force in the relevant timeframe before you scale up in your instrument. Also each position is configured specifically to the statistical data set of the relevant timeframe. This is a bit different to pyramiding on a single timeframe using retracement additions yet it allows you to make hay while the trend shines.

The method is described a bit more here :-)

ı will

The Benefit of Using Multi-timeframes for Trend Following

Below is a snapshot of 6 different time-frames applied to silver ranging from MN to M30.



Have a close look at the SDC's being applied to each timeframe.

Note that on the MN timeframe, we are just testing the short trend that has been in place since 2011. If you were a trend follower using this timeframe, you certainly would not be considering entering a position at this point and you would be waiting for confirmation that a new bull trend has commenced or that price may revert back into a short trend. You would also come to the same conclusion using the W1 timeframe as a guide.

However look what is being achieved by scaling into positions on the lower timeframes.

​While the MN chart was signalling that the short trend was being tested, the M30 timeframe was signalling that significant momentum of a bullish nature was being experienced. Now a traditional trend follower using the M30 timeframe as their guide would seriously be considering entering a trend with this string momentum drive, however for the longer timeframe trend followers, they would be sitting and waiting.......an opportunity lost.

As you can see in the chart above we have entered long on the M30 timeframe and are now deep in profit at >2.5R. We are also in good profit on the H1 timeframe. Clearly the H1 timeframe is now in a position where even in a worse case scenario, it is well beyong it's breakeven so at this point we look at the next higher time-frames (being H4 and D1). Conditions were ripe that a trend was in place on these timeframes with strong momentum, so we took entries only recently to ride the momentum drive without having significant portfolio correlation exposure.

While a traditional long term trend follower may still be sitting on their hands waiting for a bull trend confirmation, we have managed to significantly scale into a possible new bull trend that will emerge on the longer timeframes and make hay while this new opportunity opens up.

If things go according to plan we will be entering the W1 timeframe when the D1 position achieves a worst case breakeven and finally the MN timeframe. At that point in time we will have scaled in 6 x into this new bull trend and be reaping the rewards probably at the same time that a long term trend follower would be just starting to pull their trigger fingers. Our lower timeframes will be capturing the swing components of the longer term trend and hence may have several bites at the cherry as the trend unfolds and the longer timeframes will be riding the secular wave.

this is the dificult part of coding it as algorithm cuz you make decision with eyebaling and making decision.eg month chart bounce over median line and smaler tf break long