Bet on Bollinger

Ever since John Bollinger introduced his Bollinger Bands in the early 1980s they have been a favourite indicator to all technical trades. This article is about the probabilities of Bollinger bands.

How good are the chances to be outside or inside of the bands? How do these probabilities relate to the current position of the market relative to the bollinger bands? What impact has overall volatility on these statistics? These questions will be answered.

Bollinger Forward Prediction

By definition of the indicator most of of the times the market will trade inside the bollinger band. But what happens in the future? Where will the market be in some days from now. These are the questions which interest me from a truing point of view.

So I did some test on the forward prediction qualities of the bollinger band indicator. For all tests I used the 20 day, 2 standard deviations setting, which is the standard setting for most charting packages. I then looked at the positioning of the market 20 days after relative to its current bollinger band.

 

The first test was done on the S&P500 index, using data since 1983. The indicator shown on the screenshot below gives the probability of the market being outside today’s bollinger band in 20 days from now.

The red line is the probability of being outside todays band if we are already out of the band today.  The green line shows the probability of being outside of today’s band in 20 days from now if we are trading inside of the band today.

The outside starters being outside the band in the future more likely than the inside starters, is most probably due to the trendines of the market. Once a break is done there is a high probability it either vigorously reverserses or that it carries on. Both events lead to an outside of today’s bollinger band event.

100%-outside probability is the probability to be inside todays band in 20 days from now. It is less than 50%, regardless of today’s market relative to the band.

Stability of results

Continue reading

Scanning for Support and Resistance Probabilities

I have been in search for a signal I could use for a short vertical spread or naked short option strategy. So my main concern has been to find a level, which will most probably not be penetrated over the next few bars.

This is what I came up with.

Support and Resistance

We are all familiar with oscillators like the RSI indicator. It gives an idea if the market is oversold or overbought.

The chart gives a basic idea of the signal I am looking for. Once the indicator is leaving the overbought / oversold area, there should be a good chance that the market actually stays above or below it’s previous high or low. If this probability is high enough, it would be a great signal for a short vertical spread or to sell a naked put / call option. (be aware of the unlimited risk in the naked short trade!) Both strategies win, if the selected level is not penetrated at expiry.

What is manually drawn on the chart above can also be done automatically. The following chart shows how it looks like if you use the code given at the end of the article.

Every time the RSI leaves the extreme zones the indicator will draw the previous high or low for a given prognosis interval. To enhance the chances and not to get too many signals in a trending market I also made use use of the ADX indicator. So to see a signal on the chart, RSI has to leave the extreme level while ADX signals a sideway market. This should give the best signal quality.

The three signals shown would have resulted in a winning trade as the market did not cross the shown support / Resistance levels. But how does it work out in the long term?

Continue reading

Backtesting Market Volatility

If you want to predict volatility, you can place a bet on the option market. Just buy an at the money put and call, and at expiry day you will either win or lose, depending on the actual market move since you bought the straddle and the price you paid for the straddle. To put it simple, if the market moves more than you paid for the two options you will win, otherwise you will lose.

The fair price for volatility

When I look at the S&P500 I could buy or short a straddle with 16 business days until expiry right now for around 70$. That’s the implied volatility.

When I look at the standard deviation of 16 day returns, using the last 30 days to calculate it, it shows me a volatility of around 30$. That’s historical volatility.

When I use my own fair bet KVOL Volatility, it gives me a volatility of about 50$

Now I got three measures for volatility, but which one is the best prediction for future market volatility? And how big will the error (=wins and losses) be if we place this bet over and over again?

Backtesting volatility

Placing an perpetual bet on future volatility using the payback profile of a short straddle will give me an idea on how good historical volatility and Kahler’s volatility was able to predict future volatility. In a perfect world this virtual test strategy should be zero sum game; if not, future volatility is either over or underestimated by these 2 indicators.

If I know which indicator gives me the best volatility assumption, I can use this information to find out, if the current implied volatility of the market is too high or too low. Sell high, buy low…

The chart above gives you an idea on how I did the backtest. I place a weekly bet on volatility, based on a short straddle trade. So if I close outside the of the projected volatility, I have a loss. If the market closes inside the projection, I win. The maximum win will be the price of the volatility indicator at the beginning of the bet, the max loss is the point move within the week minus the price I got for volatility at the beginning of the bet.

Continue reading

Demystifying the 200 day average

The 200 day average is usually considered as a key indicator to tell if markets are bullish or bearish. But can this be proved statistically, or is it just an urban legend handed down from one generation to the next?

Let’s do some studies and find out.

The 200 day moving average

Looking at the chart of the S&P500 index and it’s 200 day average let’s me think that the 200 day average is actually a useful indicator to separate the bull and bear phases of the market. But the eye tends to see what the brain is looking for, so you might have focused on the crash 2008 and the bull market afterwards, but have ignored all these little breaches below the 200 day moving average which happened in between. As a trader I can`t make any money with knowing that there has been a long period under the 200 day average, i need to make my decision as soon as the market drops below the 200 day average.

Distribution of returns above and below the 200 day moving average

The chart above (on the right side) shows the returns distribution of 10 day returns. The green distribution represents the 10 day returns if S&P500 is trading above it`s 200 day moving average, the red line represents the 10 day returns returns when the market is trading below its 200 day moving average. Data from 1980 up to now has been used.

What are the curves telling us?

Continue reading

Seasonal trouble ahead

If a bitchy prime minister and a crazy president weren’t enough, for the upcoming months the seasonal chart is also indicating further price setbacks.

Seasonality of DAX

Analyzing the average monthly performance of the German DAX index a distinct pattern of seasonality can be observed. On average June has been down 0.6%, but the big trouble is yet to come.

 

The chart shows the average monthly performance of the DAX index, using data since 1999.

Each bar of the histogram represents a specific months percent performance. Starting with the dark blue bar in January, the green bar right now represents the average June performance.

Seasonal  and Volatility Prognosis:

As you can see on the above chart June always has had a bearish prognosis over the last years. July might bring some relief (the positive magenta bar behind June), but therefore August and September surely got a strong bearish setup. Although the markets have been bullish over the last 20 years, the average combined performance of August and September is -4%.

The average performance of a month is not a good indication for the actual magnitude of the upcoming market move, it just is an indication for its direction.

Continue reading

KVOL Volatility part 2

How to calculate volatility based on the expected return of a straddle strategy has been shown in part 1 of fair bet volatility KVOL.

Using and Displaying Volatility:

KVOL uses the given amount of historic returns to calculate an expected value of an at the money put and call option. The sum of these prices are the historic fair value for implied volatility. It can be used to compare current market implied volatility to historic fair values.

Beside calculating KVOL for a specific return period it can also be used to show it as a projection indicator on the chart.

The example on the chart gives such an expectation channel for the s&P500 at the beginning of each month. The 250 days before are used to calculate KVOL. The line underneath the chart is running KVOL for 13 trading days.

Simplified trading:

to win, with higher volatility expected: you would have bought a straddle at the beginning of the month, expiring at the end of the month. You should not have paid more than a KVOL for 25 bars (working days to expiry) would have suggested. You win if the chart is outside of the projection at the end of the month.

The shown example uses the 250 daily bars before  the beginning of the month to calculate the returns and the price of KVOL. The projected lines represent the winning boundaries of the straddle at expiry.

Continue reading

Statistics of VIX

The CBOE volatility index VIX  measures the market’s expectation of future volatility. It is the gauge of S&P500 equity market volatility.

The spikes to the top and the long phases of relatively low volatility are reflected in a left-leaning distribution diagram and a long tail towards the higher panic levels. The median value is 17%, meaning 50% of the prices are above (below) this level.

The next chart shows the distribution of returns over 25 trading days. The median price movement being slightly shifted to the negative area shows the mean reverting characteristics of volatility.

Analysing the level of VIX and the returns afterwards yields an even more interesting picture:

The green line gives the 25 bar percentage returns of VIX, with VIX noting above 25, the red line gives the returns with VIX below 15. Observe the median of the two lines:

The median 25 bar return with VIX above 25 (green) is around -15%, only 20% of the returns are positive. The return with vix below 15 (red) is above 0% and with a fat tail to positive returns. Data from 2004-2018

 

The above chart suggests that going short on volatility, if VIX is above 25, seems to be a good idea, the next chart shows what will most probably go wrong during the next 25 days. The distribution diagram gives the maximum adverse movement of the VIX.

The green line, VIX above 25, shows a +10% median maximum up movement over 25 days. So do not expect a short vola position to be without risk.

 

On the other side, the distribution of the maximum loss of the VIX during a 25 day period shows a median of below -20%.

 

Best to test for yourself,  returns distribution indicator for tradesignal code

 

Kahler’s fair bet volatility

Volatility is a measure of risk. It describes how far a commodity will most probably move within a given period of time. The most common measure for volatility is historical volatility. But I do not like the complicated formula for standard deviation.

There has to be a better way to explain and calculate volatility….

Implied Volatility

The options market has got a perfect measure for volatility. Done without formulas, just by demand and supply. And as I believe in efficient markets, the option markets fair price for volatility will be my starting point.

To get the price for volatility at the option market you just have to place  a bet. Assume you want to know the (expected) volatility for the next 30 days, then you would just add the price for an at the money put and call with 30 days to expiry. Option traders call this bet a straddle, and you would win if the market moves more than the price you have payed for the (european style) put and call.

The fair price for a volatility bet

Implied volatility and this Straddle bet is the starting point to calculate my own volatility measure.

The fair price for this bet is, when neither the buyer nor the seller of the bet has got an advantage. In the long run it should be a zero sum game game for both of them. Calculating the fair bet price for a straddle is the idea behind my volatility measure.

Think about a simple coin flip game. If you bet on head you can either win 1€ if head is up or nothing if tail comes on top. What would be the fair price for such a bet?

As head and tail got the same probability, the expected return of a bet on head’s up would be 0.5€. If I would sell you a bet on the next coin flip, I would charge you this 0.5€ to make it a fair bet. So you would either lose the 0.5€ if tail’s up, or win 1€ -0.5€ if head’s up. In the long run this would be a zero sum game for both of us. Do the same thing for the tail is up bet. It also got a value of 0.5€.

Historical Volatility vs. Kahler’s Volatility:

Historical volatility uses standard deviation of daily log returns to describe the volatility of the market. The standard deviation of this +1 -1 coin flip experiment would be 1€. The same would be true if you would buy a head’s up and a tail’s up bet; it would also cost you 1€. So for this simple example the fair bet based volatility is the same as the historical volatility.

But the market is not a coin flip. There will be some differences between historical volatility and KVOL fair bet based volatility.

KVOL vs. historical volatility:

The chart shows you a comparison between KVOL (blue) and historical volatility (standard deviation). On the chart shown above both calculate the volatility for 10 day returns, using the previous 30 bars as data sample.

As you can see historical volatility and KVOL are highly correlated.

But there are some major differences:

As an example in the end of 2017/beginning of 2018 KVOL starts to rise as the market is exploding to the upside. This is due to the virtual call used to calculate KVOL gains value. At the same time historical volatility stays low, as the market has got one direction and no setbacks.

Another advantage of KVOL is it`s response to singular events. As you can see on Sept. 3rd on the chart above the singular big red candle leads to a spike in historical volatility. It also raises KVOL, but not as much. As both indicators are calculated over the same period of bars they both got the same speed of change, but when you have a look at the scale you will see the advantage of KVOL: Historical volatility jumps from 0.2 to over 0.5 – it more than doubles just because of a single event. KVOL also raises,but only from 0.2 to 0.3.

For me this mild response to to singular events is the main advantage. Imagine a portfolio based on value at risk – would it really be useful to half the exposure just because historical volatility jumps after a single red candle?

KVOL  – Tradesignal Equilla Code:

The code to calculate KVOL is simple and straightforward.

The inputs:

multi: just a multiplier, like you can display 1 or 2 standard deviations..

datapoints: The number of bars used to calculate KVOL

returnperiod: calculate the volatility for 1,2,3… bars

showresult: show the result as a percentage of the underlying or as an absolute number

show: show either kvol or the rank of  kvol within the last 100 bars. This gives an idea if volatility is high or low

Meta: subchart(true);
Inputs: multi(1.0), Datapoints(30), returnperiod(5), showresult(percent, absolute), show(result,rank);
Variables:Kvol, i, rp,rc, rpsum, rcsum, call, put, hh,ll;

rpsum=0;
rcsum=0;

for i =0 to datapoints-1 begin // loop over last bars
  rc=maxlist((close[i]-close[i+returnperiod])/close[i+returnperiod],0); // % return of call
  rp=maxlist((close[i+returnperiod]-close[i])/close[i+returnperiod],0); // % return of put
  rcsum=rcsum+rc; // sum of all %returns over time
  rpsum=rpsum+rp;
end;
      
call=rcsum/datapoints;
put=rpsum/datapoints;

Kvol=call+put;	
if show=result then drawline(multi*iff(showresult=percent,100*Kvol,Kvol*close),"KVOL");

hh=highest(kvol,100);
ll=lowest(kvol,100);
if show=rank and (hh-ll)>0 then drawline(100-100*(hh-kvol)/(hh-ll),"rank");

 

keep researching…

 

 

A graphical approach to indicator testing

The first step in algorithmic strategy design usually is to find some indicators which give you an edge and tell you something about tomorrow’s market behaviour. You could use a lot of statistics to describe this edge, but I like to take a graphical approach in indicator testing first, and only later on worry about the math and statistics.

Scatter Charts

A scatter chart is a simple to read chart style to see the correlation between two input values. A regression line on the scatter chart gives you a visual idea if the two securities are positively or negatively correlated, the “cloud structure” of the scatter points tell you if this correlation is tight or loose.

This sample scatter shows the correlation between the DAX and DOW levels, and it can be easily seen that these two markets are tightly correlated in a positively way.

The horizontal scale is used for the second security (dax), the vertical scale is used for the first security (dow). This chart type is predefined in Tradesignal, just drag&drop it onto the securities on the chart and select the right amount of data to get the analysis you want to see. (eg. 2000-now). If you see a tight and positive correlation like on the chart above, It migt be used to select the instrument you want to trade. If market A is easier to predict than market B, select A.

Scatter on Indicators

Although a scatter chart is usually used to show the correlation between two markets, it can also be used to show the correlation between two indicators.

The chart above shows the correlation between digital stochastic and momentum. Have a look at the clustering of points in on the right side of the scatter, a high level in digital stochastic usually goes with a high momentum. This insight enables you to get rid of momentum, as digital stochastic is easier to read than the shaky momentum. Less indicators = less parameters = less curve fitting.

Scatter prognosis

Doing this analysis and getting rid of parameters is great if you want to minimize the dangers of curve fitting, but it does not tell you if your indicator is of any use at all, when it come to describing tomorrows move of the market. Surely it is valuable insight that a high level of stochastics corresponds to a high momentum, but does a high momentum today also mean that the market will move up tomorrow? And this question about tomorrow is the key question I ask myself when searching for some edge.

To get a glimpse on the prognosis quality of an indicator we will have to add some colour to our scatter chart. This colour tells me what the market has done after a specific indicator level has been reached. Green for an up move, red for a down move, black for not decided by now.

This chart shows the prognosis quality of the stochastic indicator. The left chart shows the 1 day prognosis of a 5 day stochastic, the right chart gives you the 5 day prognosis of a 21 day stochastic. Observe the clustering of the red and green dots. (black for not decided by now) As you can see on the left chart, the one day prognosis using a 5 day stochastic is not the thing to do. Regardless if stochastic is high or low, you get a nice mixture of red and green dots. This means the market, at a given stochastic level, sometimes moved up, sometimes moved down. Not this behaviour is not very useful for trading. Only in the extreme, near 0 and 100, this indicator seems to implicate a bearish next day movement.

The right chart, showing the longer term prognosis of a long term stochastic seems to be more useful. High levels of the indicator also show positive returns on the 5 days after, unfortunately you can not reverse the logic, as low indicator levels give a rater mixed prognosis. This visual analysis can give you an idea which areas of the indicator might be useful for further analysis.

A one dimensional analysis like on the chart above could also be done without this scatter chart. Going from one dimension to two dimensions is more useful, as it directly can be translated to do a kNN machine learning trading strategy. Have a look at the following chart. It shows the scatter of two indicators and the implication on the next days market move.

Lets start with he right chart. As you can see the red and green dots are evenly distributed, meaning there is no useful correlation between the used indicators and the movement of the market on the day after. If you would use a kNN algorithm with these two indicators, I would bet it would not return great results. Even if you would get a positive return, it might just be a lucky hit or curve fitting.

The opposite is true for the chart on the left. Here you can see some nice clustering of the red and green dots. Low indicator levels seem to predict a bearish move, high indicator levels result in a bullish move on the next day. A distribution like this is the perfect starting point for investing some time in a kNN machine learning  trading strategy. The kNN algorithm would give you a strong prognosis with high or low indicator levels, and most probably only a weak or no prognosis when the indicators are around 50. The returns will be stable, no curve fitting problems should be expected.

Conclusion

Using a scatter chart can give you a nice visual indication if your indicator might be useful for prognosing the next days market move. This is valuable insight, as you can see the whole data universe with one glimpse, even before you do a thoroughly statistical analysis. Numbers can deceive you, pictures usually tell the complete story.

Tradesignal Equilla code:

 

 

 

Machine Learning – KNN using Tradesignal Equilla

I always thought that inspiration and experience is a key factor in trading. But everytime my chess computer beats me without any inspiration, just by brute force, I start to reconsider this assumption. This article will be about a brute force approach in trading.

Rule based trading

I have never been a great believer in classical technical analysis. If you ask 2 analysts about the current trend in the market, you get at least 3 answers. So I turned to algorithmic trading, using the tools of technical analysis in a new way, doing if..then conditions, backtesting them, refining the rules and parameters until the desired result was shown. These if..then based conditions, like if the market is above it`s 200 day line then go long, are mostly found by experience and inspiration. Isn’t my brain just a neural network which can be trained with historic data (experience), enhanced with a glass of wine for the inspiration?

Today I would like to take a voyage into machine learning. I would like to let my computer find the rules by itself, and just decide if I like the results or not. I’ll have the glass of wine with some friends and let the machine do the job; This sounds tempting to me, but can life really be as easy?

Unsupervised machine learning – kNN algorithm

The knn algorithm is one of the most simple machine learning algorithms. Learning might be the wrong label, in reality it is more of a classification algorithm. But first let’s see how it works.

The scatter chart above is a visualization of a two dimensional kNN data set. For this article I used a long term and a short term RSI. The dots represent the historic RSI values. Have a look at the fat circled point. It just means, that todays RSI1 has a value of 63, and RSI2 got a value of 70. Additionally the dots have got colours. A green dot means the market moved up on the following day, a red dot shows a falling market on the day after.

kNN – k nearest neighbours

To do a prediction of tomorrow’s market move, the kNN algorithm has a look at the historic data (shown on the scatter plot) and finds the k nearest neighbours of today’s RSI values. As you can see, our current fat point is surrounded by red dots. This means, that every time the 2 RSI values have been in this area, the market fell on the day after. That’s why today’s data point is classified as red. Call it classification or prediction, the kNN algorithm just has a look on what has happened in the past when the RSI indicators had a similar level. Have a look at this video, it is a great explanation on how the algorithm works.

kNN as Tradesignal Equilla Code

Computer kiddies would implement this algorithm in Python or R, but I would like to show you an implementation with the Tradesignal programming language Equilla. It is way faster than Python, and has got the advantage that I can directly see all the advantages and disadvantages on the chart. It is not just number crunching.

To implement the algorithm in Tradesignal we first have to do the shown scatter plot, but not graphically but store the 2 rsi values and the next days market move(colour of dots) into an array.

In line 8&9 the rsi values are calculated, line 12&13 calculates the next day`s market move. Line 15 to 20 then stores the data into the training data array. This is done for the first half of the data set, for my example I will use the data from bar 50 to 1000 on my chart of 2000 data points.

The next task to complete is to calculate the distances of today’s rsi point to all the historic points in the training data set.

Line 23 to 27 calculates the euclidean distance of today’s point to all historic points, line 29 then creates a sorted list of all these distances to find the k nearest historic data points in the training data set.

We are nearly done. The next step is just to find out what classification (colour) the nearest points have got and use this information to create a prediction for tomorrow. This is done in lines 33 to 35

Have a look at the scatter chart at the beginning. If this would be the data stored in our training data set, the prediction, using the 5 nearest neighbours, would be -5. All the 5 nearest neighbours of our current data point are red.

Now that we got a prediction for tomorrow, we just have to trade it:

kNN algorithm performance

Lets have a look if this simple machine learning algorithm works. Using 2000 days of backward adjusted brent data, I used a 14 and 28 day RSI to predict the next day move. The training was done on bar 50 to 1000, and I used the 5 nearest neighbours for the classification.

Knn algorithm – conclusion

Judging on the shown graph it seems to work. It seems to be possible to use these 2 RSI indicators to predict tomorrow’s brent move.

kNN algorithm gives me a framework to test all kind of indicators or even different data sets easily and see if they have got any predictive value.

This is definitely an addition to classical algorithmic trading, using if..then conditions build from experience and intuition.

But you might still need some intuition to find the right data sets, indicators and parameters to get a useful prediction. Not everything can be done by machine learning…