mindmoneymarkets

Trend following & when the market won't pull back.

[I wrote a draft of this piece the other night and with well timed irony, the S&P 500 has finally pulled back]

One thing that's puzzled me is what to do with trend following markets. There has been a lot of research on what to do with a mean reverting market using short term indicators such as the DVB or RSI (2) (2 period RSI) - I.e. buy over sold and sell over bought readings etc. 

But what about markets which are not as prone to mean reversion? Should you just reverse the rules and buy on strong readings etc? Another options its to employ a relative strength rotation model which holds the best trending markets as with the Livermore Active Issues. 

Many people have also be commenting on the recent rally on the S&P 500 and other markets. I can't remember the exact stats, but Bespoke pointed out that it had been some time since there was a 1% correction. 

Anyone trading the basic short term mean reversion rules for indicators such as the RS2 or DVB will have had their patience tested in recent weeks as the market continued to push higher. The screen grab below shows the S&P 500 with the DVB indicator below it. The basic trade idea is to buy when the DVB drops below 50 and sell when it moves above 50. As you can see, the DVB has been above 50 for some time. 

I wondered what would happen if you used the average of the DVB as a trading input, specifically to adapt any methods to situations like this. 

I tested the 20 period average of the DVB on the S&P 500. 

Interestingly, your average return per trade is increased by trading in line with the 20 period average of the DVB. For example, if the DVB is below 0.5 and the average DVB is above 0.5, no trade. 

It's not a spectacular addition, but maybe worth some further investigation. Looking into the standard deviation of the DVB or RS2 also seems to be useful from my quick study. Essentially the returns from trading a short mean reversion strategy are lower when the standard deviation of the short term indicator is below average. This kinda makes sense as if you look at the chart above, you can see the DVB barely moving for weeks, which is what you get during a steady uptrend (bad for mean reversion). 

Trend following markets

This got me thinking about trend following markets and the best approach to take. If a market is more prone to trend following then trading short term mean reversion will be ineffective and trading the opposite rules may actually work out better. 

Foreign exchange (Forex) is certainly more prone to trend following and in my experience short term indicators such as RS2 don't work when applied in the usual way. They actually make a small profit when traded with the opposite rules (Buy on high readings etc). 

I wondered what would happen if you applied the 20 period average of the DVB to EUR/ USD (Euro to US dollar spot rate). The red line is the 20 period moving average of the DVB. 

 
The rules were as follow: 

Buy the EUR/ USD when the 20 period average of the DVB is above 0.5
Sell(short) the EUR/ USD when the 20 period average of the DVB is below 0.5. 

No commission spread etc. EUR/ USD spot since 1998. 

The average margin per trade is small, with the average trade making 0.03% profit with a 51% strike rate. Too small in its own right to be a devastating strategy. However, it does lay out a good platform on what is usually a very tricky market to trade and I've found a few interesting angles already by delving deeper. I'll have to keep those under my hat for the moment though. 

Some thoughts: 

• There is a massive difference in the reaction of the euro ETF (FXE) and the EURUSD spot rate in how they respond to mean reversion strategies despite being highly correlated, they appear to be two different beasts and this difference mainly came about in the last 12 months. 
• I have only quickly tested the RSI2 average and it appeared to make similar ish results, though not as good. 
• Standard deviation of the DVB/ RSI2 etc could be fruitful area to look at. 
• Correlation between the EUR/ USD and the SPX has also yielded some interesting results for the average DVB idea. As the SPX has responded well DVB type strategies in the last few years (though as MarketSci point out, simple buy on down day type stuff has gone off the boil), this will also impact on currencies when they are correlated highly with stocks. 
• The dollar index actually performed the best in my testing compared to the eur/ usd, indicating that it is the dollar itself on a trade weighted average which is the core trend following element of the eurusd (though the euro contributes a big slug of this trade weight). 

Appreciate any thoughts...

[Post Script] David Varadi's De trended indicator concept looks to be an elegant way to deal with trending markets. I just haven't had chance to analyse it yet. 

Position sizing with TVA

I've been playing around with the Targeted Volatility Analysis ideas presented on CSS.

Some interesting results.Check the link for more details, but basically TVA adjusts position size according to a relatively simple volatility calc called TVA. You can apply this to a market like the FTSE or in theory a trading system.

I'm not sure I've got the calc a 100% right, but here's the results of trading the FTSE 100 with it.

Note I've used it more of binary indicator which isn't the intention, but in the real world unless you are an institution, you're not going to be able to adjust your market position every day. So the TVA below 1 basically gets you out of the market. TVA of one means you stake 1 X the TVA. I didn't double exposure beyond this.

It does quite a good job:

 

The worst drawdown is cut to 1700 index points from 2930 index points.

I then wondered what would happen if you look at an indicator like the RSI2 period and traded this long only when you also got the go ahead from the TVA.

Rules: Buy only if the 2 period RSI is less then 30 and the TVA is above 1. Position size is 1 X the TVA.

 

The filter adds quite an improvement to RSI2 on a long only basis with the staking sizing helping to increase profits.

Unfortunately it hasn't helped the RSI2 method make a profit in 2010, but the losses have been marginal.

I've also combined this with an additional EV filter, but I'll put that up another day. Hope this is of interest.

A big thanks goes out to Jerry and everyone involved for sharing such an interesting idea.

Where's the action?

As an extension of things I've been looking at in relation to mean reversion, time zones & ETFs, I thought it would be interesting to look at where the action has been happening on stock markets in the last few decades.

By this I mean, do you get bigger gains between the close of the previous day to the close of today? Or does the action all happen between the open and the close?

To examine this I looked at three scenarios on the SPY (S&P 500 ETF).

Before I continue, I must stress this is nothing new and I'm sure Michael at Market Sci explored this (apologies if it was someone else, I can't find the link) 

On the SPY, going back to 1993, I compared

A) Buying at the open, selling at the close. O to C
B) Buying at last night's close, selling at today's close. C to C
C) Buying at last night's close, selling at today's open. C to O

I've ignored any practicalities such as being able to actually trade exactly at the open. 

Here's your results:

 

The red line (Close to close) is effectively the SPY, though due to the way of totaled the % each day it's not an exact match. Pretty close though.

The most striking finding is that buying at the open and selling at the close produces very little by the way of gains or losses. The bulk of the gains happen over night, not during the day's session. It's important to stress I'm talking gains here by effectively being long. If we were to examine the absolute moves, there would be different findings.

Next, just trading from the last nights closeto the open today has out performed holding until the close. Quite amazing really.

Is this just a SPY phenomenon?

It appears not.

I performed the same test on the ETF of the FTSE 100 listed in the UK, the ISF. I could only go back to 2002 when the ETF launched. I can't use the FTSE data because its opening and levels are generally based on the previous close, whereas the ISF are generally more realistic like the SPY vs the SPX.

   

Again, buying at the close and selling at the open of the next day out performed close to close trading while for the most part, buying at the open and selling at the close under performed significantly.

Can we trade this?

I'm not putting this forward as a system as it has been made with pure hindsight, but I thought it was interesting all the same.....

If the day's gains tend have already happened by the open, could you profit from shorting the open to the close after an overnight gain? Well this on its own isn't enough so I added what I think are some fairly logical filters.

These measure:

1. The 20 period percent rank of the absolute close to open move size. (Effectively an average move filter)
2. The 252 period Percent Rank of the close to close total performance. (Effectively a trend filter)

Go short at the open until the close if:

1. The move from last night's close to the open was positive.
2. The 20 period PR of the absolute close to open moves size is below 0.75 (i.e. it wasn't a huge move up overnight).
3. The 252 period PR of the close to close total is below 0.8 (i.e. a down trend)

In plain English, you're shorting overnight strength during a down trend provided the move isn't too big.

How does it perform?

Pretty well on the SPY Considering you're only really trading during down trends.

Incidentally the 26th would have been the 1st trade for a while, but a 1.2% gainer shorting the SPY.

The model doesn't out perform on the ISF as much, but I reckon if the data went back to pre 2000, it would show a similar effect as the SPY (in other words, its missing one major down trend that makes it work). The data only goes from 2002.

Food for thought anyway.

DV Indicator system filters

Over the last few days I've been trying to get to grips with DV Indicator's excel plug in. Excellent stuff in there but it's taken me a while to unwrap the parcel so to speak.

Getting there now and have been experimenting with some of the filters with some interesting results. The plugin essentially combines most of the research DV has been putting up on his blog, allowing you to do some basic back testing of each indicator and create panels which are a springboard for your own research.

It's the latter I've been playing with recently. I especially like the Zones filter which allows you to determine if the market is in one of 6 zones from uptrend with low vol to down trend with high vol.

I had a play around with some indicators on the ISF (FTSE 100 ETF) and SPY (S&P 500). Here's what I've got so far. Not particularly psyched about this 'system', it was more to try out the features. To big a portion of the returns still come from the mean reversion frenzy period in 2008.

Main indicator: DVST - A trend indicator
Filter 1 = DVB - The DV2 bounded indicator
Filter 2 = DVPV - Percent rank of volatility
Filter 3= DV zones

I like the way the zone filter (purple) allows you to cut out a bunch of losing trades and keep total returns ahead of the market. 

The S&P 500 appears to be more receptive to these sort of filters which is no surprise:

No zone filter with the SPY yet.

Anyhoo, it's an interesting start.

1st Day Of The Month Strategy on the FTSE 100

I've blocked a few days out to get to grips with DV indicator's excel plug in, but I couldn't resist doing some quick UK focused version of some of the seasonality stuff that MarketSci has been exploring.

Specifically I wanted to see if the 1st day of the month effect held true for the FTSE 100.

Turns out it does and works even better with a simple moving average filter.

Data: FTSE 100 back to 1984
Rule: Go long on the close of the last day of the month. Exit on the close of the 1st day of the month.
Filter: 50 period moving average. Only go long if the FTSE 100 is above its 50 period Moving Average.
I also had a look at other filters (200 MA, 252 Percent rank which also improved returns).

How does it do?

Pretty well!

Here you can see results of the method compared to the results with a filter.

The next chart shows all the various filters and the FTSE 100 returns since 1984.

 
The longer term filters (200 MA, 252 PR) have a lower return, but have a slightly lower Standard Deviation.

My take

I wouldn't dream of trading this on its own, but I do find it incredible that with the 50 MA filter, there have been just 200 trades since 1984 vs nearly 6,600 trading days for the FTSE 100 in total. Yet these 200 trades (3% of the total available trading days) have produced over 75% of the profits. Even without the filter, if you take out the 1st day of the month, your trading returns would be reduced by over half since 1984.

Fascinating.

There are probably many logical and nonsensical reasons why this anomaly exists, but at the end of the day, it's impressive that it's so well known and survived so long. Perhaps it's the hot anomaly momentum becoming a self fulfilling prophecy. CXO presented a good strategy on switching between these hot anomalies and it may be worth investigating a few of those presented further.

Download 1DOMFTSE

The real cost of trading part II

My previous post showed the results from trading a system with a very small edge. It revealed that if the differential between your average winner and your average loser is small then a wide spread will kill your profits.

But what if you've got a healthy margin and a healthy strike rate?

Let's have a look...

Here's the assumptions I've made about costs.

Costs	Spread	Commission	Other	Assumed level	Unit	Multiplier
Spread betting	1			1000	1	50
Futures	0.25	$5.32		1000	1	50
SPY	0.025	$5.00	0.30%	100	0.1	500
Fideltiy Funds Net	0.025	$5.00	0.50%	100	0.1	500
Fideltiy Funds Net (No fees)			0.50%	100	0.1	500

Same as the previous post before mostly. Market is S&P 500, emini s&P 500 or SPY or equivalent from Fidelity funds network.

Note with regard to the Fidelity funds network, this is a hypothetical comparison because the amount you can invest within the ISA wrapper is limited + you cannot go short + I have not investigated the tracking error of the S&P 500 equivalent funds. Fidelity funds network has no switching fees or commission on some funds. I've used this as the frictionless scenario.

There may be similar schemes available with a SIPP, but I haven't checked. 

Finally, as far as i know, Capital gains tax is due at 18% on realised gains over £10,100 in the UK. Some tax advisors say this should be taxed as income, depends who you speak to who the day of the week it seems.

CGT is liable on Shares & futures but not spreadbetting or fidelity funds network ISA wrapper.

The data does not include dividends.

Scenario 1:

Trades: 200
Winners: 124
Losers: 76
SR: 62%
Ave. Winner 2.83%
Ave. Loser -2.22%

NB these are the results I have for trading the AMI system on the SPY since 1997. Long >0.5 and short <0.5.

 

You can see that the wider spread on spread betting reduces returns, but you gain a lot by avoiding CGT. So if you have a reasonable margin on your trades then spread betting is worth it, but only because of avoiding CGT.

Naturally if you expect your annual realised gains to be closer to the CGT threshold, then you might want to consider the reduction in returns you'll feel from the wider spread. Many traders would be happy with £10,000 profits each year so it's something to consider. Naturally futures comes with a higher cost of entry with the emini S&P 500 have a $50 per point multiplier. 

Perhaps you could trade with CGT liable methods until you hit the annual threshold then switch to tax efficient vehicles?

NB I've ignore $/£ differences here. The example above is for comparison only.

Scenario 2:

Trades: 200
Winners: 150
Losers: 50
SR: 75%
Ave. Winner 1.00%
Ave. Loser -1.50%

Here I'm looking at the results for a system with a high strike rate, but with a the average winner being bigger than the average loser. It still makes a profit due to the healthy strike rate, but there's not much margin for error on each trade.

Here we can see returns are better with low bid/ ask spread vehicles even when taking into account CGT. I see many systems marketed in such a high strike rate/ low ave trade ratio and naturally none of them take into account the real cost of trading such a method.

More to come...

Anomaly/ Regime momentum

After my recent post on indicator rotation, I noticed the excellent CXO advisory put something up on anomaly rotation.

Their general finding is that investment anomaly's do persist and basing your trading/ investment strategies on the hottest anomalies can produce outstanding returns.

What is an anomaly?

Well according to efficient market hypothesis, there should be no profitable angles that maintain over time. Something simple like trading around the January effect (stocks out perform in January) shouldn't work in the long run because in theory, the moment something starts to work, the market starts to discount it, thus neutralizing its effect.

Except, it seems that the effect does persist for longer than it should according to the Efficient Market Hypothesis. Hence the term anomaly. Perhaps for a while the effect persists because enough investors buy into it, thus making it a self fulfilling prophecy. Perhaps there are underlying reasons for the anomaly that persist for logical despite their existence becoming widely known.

'Trading regime' could be another way to describe this phenomenon. Mean Reversion is perhaps a good example of a trading regime. As you can see with this post from Market Sci, Mean Reversion is currently a dominant regime/ anomaly but that's only been the case since 1997 until then, the market experienced more follow through momentum. 

The are perhaps logical reasons for mean reversion. The average period of stock ownership is now measured in days not years so perhaps traders are more prone to take profits quickly after an up day and vice versa. Who knows, perhaps its best not to over analyse these things. About the only thing we can say with reasonable certainty is that mean reversion is an anomaly that has persisted.

It won't do so forever.

Short term indicator rotation

So in effect the last post about indicator rotation is looking at persistence of various sub themes of an anomaly (mean reversion). It seems that these sub themes (RSI2, Dv2 etc) may have their own momentum, just as regimes themselves have momentum. 

By implication, you could potentially build an adaptive system that A) trades the anomalies with the best momentum B) trades the best sub themes of those anomalies. 

This is starting to sound like Star Trek.

CXO have whole series of posts on this which I'm working my way through including:

• Success of switching across 'style' ETFs based on 6 month momentum.
• Beat the market with hot anomaly switching.
• Diversifying across anomalies

Short term indicator rotation model

Following on from this post on the best Short term indicators, I wondered if would be possible to create a rotation model on indicators just like you would momentum stocks. I.e. you back the to performing indicators only each month/ week/ year and rotate your strategies accordingly.

Well from my small sample size of three systems (all of which are very similar it must be pointed out), it turns out that this idea might have potential.

I did a sum total for each indicator (DV2a, Dv2b, Ultimate Oscilator and the market itself, SPY) for each month.

I then took a rolling total of 5 month returns from each indicator + the market. Follow only the indicator with the highest total over the last 5 months for the next month. Repeat process at the end of every month. If the market is the top performing activity then stick with that.

All indicators set to >0.8 as the sell trigger and <0.2 as the buy trigger. No commissions/ spreads - I'm looking for a proof of concept at the moment not a trading system yet.

Here's the rolling total monthly returns. The 5M Av is the rotation model (the 2nd best line if you can't see the colours).

This doesn't tell the full story though. I attempted to show the risk adjusted returns by comparing the standard deviation to the total returns. I then divided the returns by the standard deviation. SD here is the raw monthly standard deviation, not a % of the SPY's value.

	DV2(a)	SPY	DV2(b)	UO	Rotation
StDev	4.213346	5.745971	5.839076	4.70817	5.988967
Return	160.59	36.44	282.98	218.78	274.25
SD adj Return	38.11	6.34	48.46	46.47	45.79

Again this doesn't quite tell the whole story though because the Stdev measure used can penalise a system for upside volatility. So let's have a look at downside standard deviation only. 

Negative SD only	DV2(a)	SPY	DV2(b)	UO	Rotation
StDev	2.570973	3.975319	2.723411	2.529702	2.323051
Return	160.59	36.44	282.98	218.78	274.25
SD adj Return	62.46	9.17	103.91	86.48	118.06

Now we can see the real potential with the model, it reduces downside volatility while keeping returns steady.

Next I wondered what would happen if you hedged your bets by backing the top 2 performing systems next month. This might mean investing fully in the SPY for one half and trading Dv2 for the other half of your funds.

Interestingly, returns are again below the top performing indicator, but this time have a much lower Standard deviation even before we isolate downside volatility.

	DV2(a)	SPY	DV2(b)	UO	Rotation
StDev	4.213346	5.745971	5.839076	4.70817	4.479649
Return	160.59	36.44	282.98	218.78	252.73
SD adj Return	38.11	6.34	48.46	46.47	56.42

Here's the hedged version with downside volatility isolated. 

Negative SD only	DV2(a)	SPY	DV2(b)	UO	Rotation
StDev	2.570973	3.975319	2.723411	2.529702	1.822053
Return	160.59	36.44	282.98	218.78	252.73
SD adj Return	62.46	9.17	103.91	86.48	138.70

Here we can see that the monthly standard deviation drops markedly while keeping returns pretty constant. The risk adjusted return therefore really stands out.

So encouraging signs that this idea has legs. For the record the most common activity used was the SPY itself. 

No doubt there is a more sophisticated relative strength/ momentum model than the one I have used that could be worked out.
Would the model still work with other indicators? (non mean reversion for example?).
What if you throw in some indicators that have since stopped working (is there a survivor ship bias here?)

Lots of questions, but at least it seems like this could be a fruitful line of research.

Perhaps the most encouraging thing I've seen is this graph from the FTSE 100. Same settings as SPY. The short term indicators just don't seem to work as well on the FTSE 100, perhaps because of the delay in reacting to US markets. However, the rotation model still managed to pick out the one indicator that was working and rotate out of that when the others/ the FTSE started to improve.

Best short term indicators

There have been some great posts on short term mean reversion indicators. A recent one here from Market Sci looked at various 2 period indicators, while Trading The Odds looked at a few more under slightly different settings.

In a past post I wondered if you could improve the AMI by using a different MR component. The first task is therefore to findout which MR is best. The guys mentioned above have already done the lions share of the work so hat tip there.

While doing this, I wondered, can you apply a relative strength tracking filter to various systems in a sub category and switch between them depending on progress. You can see the below that different systems responded to the market at different times, switching between them in some way could improve returns - could

Anyhoo, I only got part way through my list, but here's the findings so far for

DV2(a) - Buy/ Sell at close of next day
DV2(b) - Buy sell at close of that day
Ultimate Oscillator

First, here's the results if buying >0.5 and selling <0.5

 

	spy	DV2	DV22	UU
CAGR	3.24%	12.70%	14.88%	9.42%
Max DD	56.65%	12.89%	22.08%	27.46%
Return	51.32%	373.37%	506.65%	222.24%

First, here's the results if buying >0.8 and selling <0.2

 

	spy	DV(a)	DV2(b)	UU
CAGR	3.24%	9.18%	12.74%	11.03%
Max DD	56.65%	13.86%	12.70%	15.36%
Return	51.32%	213.07%	375.45%	289.64%

You can hopefully see how the different indicators got there groove on at different times. Even the different DV2 options performed better at different times.

Can you perform a rotation model on systems/ indicators just like you can with momentum stocks? Who knows, I'll try to find out.... Have a good weekend.

Pairs trading large cap small cap

I was interesting in this post from Market Sci on the size effect (Small caps outperform large caps in the long run). 

The post was mainly focused on isolating the size effect.

The method was:

50% long small-caps and 50% short large-caps, to isolate just small-cap outperformance

Then:

But next let’s imagine that we instead ran two portfolios. In the first, we only took the pairs trade when the market (average of large and small caps) rose, and in the second only when it fell. Remember, we’re just looking at outperformance so the direction of the market shouldn’t matter.

He concluded:

It’s pretty simple: because small-caps are pretty well correlated to large-caps but about 30% more volatile, when the market goes up they tend to go up more (and vice-versa). So because of the general upward bias of the market, small-caps have naturally tended to outperform large caps. That’s NOT a size effect, that’s a volatility/beta effect.

I wanted to see if I could replicate the results and see if there was something tradeable and also if they help true for UK markets. 

Replication

Firstly, I used yahoo data not adjusted for dividends and could only go back to 1990. So there would undoubtedly be some differences. I didn't account for trading costs either. Also I just used the straight indices for performance data so for 50/50 split it's based on index values not account balance or account allocation. This again may explain some differences.

Large caps = S&P500
Small caps = Russel 2000

The huge difference is that I also looked for short trades.

So....

If today's average of the SPX & RUT was above yesterdays, go long the Russel 2000 and short the S&P500
If today's average of the SPX & RUT was below yesterdays, go short the Russel 2000 and long the S&P500

Anyhoo, here's what I got. 

 

One thing that struck me about Michael's analysis was that it was tapping into the Mean Reversion trend that has been so strong in recent years. Buying after a down day and selling after an up day has produced very good returns in the last decade (apart from a wobble last year).

I wanted to see how the individual components of the pair trading strategy performed ie, if we stop it being a pairs strategy and just look at the MR, how would things have gone?

Let's see.

Using the average of the RUT and SPX still, but only trading one of those markets.

 

Here we can see the MR (Mean Reversion) quick in action on large caps. The effect is different to just looking at the SPX or RUT in isolation because we're looking at the average of the SPX and RUT still to define an up day/ down day.

 

Wow, that's quite a turnaround for MR.

Small caps were the king of the trend until the turn of the new decade, since then doing the opposite and buying after a down day would have been very profitable. The middle of the chart is roughly 2003 and we can see since then that MR went nuts in the last few years for large caps.

More to come tomorrow