Blog / Position Size Selection
Position Size Selection
In this writing I try to highlight some aspects of position size.
This is Not equal to Position Management before and during market participation.
It might be quite different from person to person and strategy to strategy and time- to time-frame how we build up and tear down our positions.
As a predictor player having overnight positions I found for this purpose to build Up the position mostly in one buy / sell order and in less than 10 % of the cases in two orders.
Having a position overnight I found the best results could be achieved by closing that position in two orders, or sometime (Less than 15% of the cases in three orders) and occasionally in one order, especially if need to act quickly to limit losses occasionally.
This article makes a few considerations about possible factors, that impact our decision making, when we think about position size, before opening a position as a result of our analysis considering predictor data and other market analysis.
Position size should reflect our confidence in our trading decision.
It can have multiple factors or aspects, but best if we try to stay objective, depend on available data, though we might also add psychological factors, like our perceived mental strength at the time of position opening, especially if we are daytraders.
The available funds for market participation is our starting point and also impact the risk, that we accept to bear.
From this two we can usually get two numbers:
1. The Maximum position size, that we would take in the best case. (MaxPos)
2. The Minimum position size, that we would take in the least confident situation.
(MinPos)
Most of the time our actual position size will be between MinPos and MaxPos.
We introduce PSC, Position Size Coefficient, that will be a measure of our actual position,
Between MinPos and MaxPos.
PSC can be the result of an equation, that takes into account the different position-size related parameters that we consider important. We will detail a possible construction of PSC later.
No matter what are those parameters, PSC will have a Maximum value, PSCMAX in the best situation and a Minimum value PSCMin in the least confident situation.
See the chart below for illustration:
.
With the introduction of the above the Actual Position Size (ActPosSize) could be calculated something like the following:
To visualize it lets say we have a Minimum Position Size of MinPos $ and a Maximum Position Size of MaxPos $.
With these data I created the following chart to demonstrate visually the Actual Position Size Selection.
ActPSC, the Actual Position Size Coefficient could be calculated, taking into account the most important parameters.
The following is only one example, that those, who rely on the predictionwizard generated win probabilities for the Daily and / or Weekly predictions of a specific Index:
ActPSC = A * B * C * D where
A: Prediction Win Probability related parameter
This parameter is related to either the Daily prediction win probability Or the Weekly prediction win probability or a combination of both, depending on our preference and time – frame, that we intend to play.
A (Actual) = A(Min) + (A(Max) – A(Min)) * (Pwin – Pwin(Min)) / (Pwin(Max) – Pwin(Min))
Where:
Pwin is the actual win probability, calculated by the predictor system (It is the Daily win probability if we use only that in our calculation, but we also could use the weekly and / or the monthly predictor generated, actual win probability if we trade different time frames.)
Pwin(Min) is the Minimum win probability that is generated by the system and we use for our trading purposes.
Pwin(Max) is the Maximum win probability that is generated by the system.
A(Min) is the Minimum parameter data that we would like to get (It is our decision) for this parameter.
A(Max) is the Maximum parameter data that we would like to get (It is our decision) for this parameter.
If we would like to have the “A” win probability related parameter to have a bigger impact on the ActPSC (Actual Position Size Coefficient), than we select A(Min) and A(Max) in such a way, that A(Max) / A(Min) ratio is bigger, and if we would like the win probability related parameter to have a smaller impact on ActPSC, than we compress that A(Max) / A(Min) ratio, and select that to be smaller.
The following chart try to show the two scenario:
Note that we could select A(Max)- A(Min) range in such a way, that the medium of the range is equal to one, and when A = 1 than we can say that the Actual prediction win probability will not alter the final position size coefficient in this calculation sequence.
It is easiest to select A(Min) to be one, and A(Max) to be slightly bigger than one.
Lets say we would like the win probability - related parameter alone to have a 50 [%] impact on the ActPSC, meaning that this parameter alone might alter ActPSC by 50 [%], than we select A(Min) and A(Max) so that A(Max) / A(Min) = 1.5
So for example we can set A(Min) = 1 and A(Max) = 1.5
A will be A(Min), when the predictor win probability will be minimum (Say 55 [%]) and A will be equal to A(Max) when the actual calculated win probability of the system is equal to the maximum (Pwin = Pwin(Max) Say it is 83 [%] for the NASDAQ Daily prediction (We can get this from the downloadable prediction history data for all indices, which is available in Excel spreadsheet format.)
Note that this is the same transformation function, that we used to transform one specific data range into another at the beginning of this article.
But in our case, we have four parameters, A, B, C and D multiplied by each other, so the final impact on ActPSC will be the multiplication of each individual parameter, A, B, C, and D so if we want to decrease the impact of any one of them, than we need to transform that specific parameter data range into a smaller range.
B: Trend Direction related parameter
This parameter takes into account the information, that the predictionwizard works in such a way, so that the actual win probability, will slightly depend on the trend direction.
If predictionwizard predicts a direction and that direction is the same as the current short – to mid term trend direction (10 Day to 3 Month), than the Actual Win probability is slightly higher than the calculated win probability and If predictionwizard predicts a direction which is the opposite of the short – to mid – term trend direction, than the Actual win probability is slightly lower, than the calculated one, presented on the WEB – site.
(The calculated win probability does Not take into account the actual short – to mid term trend direction, it is an aggregate, taking into account multiple years of data, regardless of trend direction.)
So with this we can calculate B as follows:
B = ( 1 + Kt) where
Kt is a constant that we select. For example from the 0 … 0.5 range.
If the predicted direction is the same as the short – to mid term trend direction, than Kt is greater than zero.
Kt is zero, if the predicted direction is the opposite of the short – to mid – term trend direction.
For example If Kt = +0.3 then B = (1 + 0.3) = 1.3 (Predicted direction is the same as the trend direction.)
If Kt = 0, then B = (1 – 0) = 1 (Predicted direction is opposite of the trend direction.)
In this case the maximum impact of “B” for the final Actual position size coefficient (ActPSC) is: (1.3 / 1 * 100 – 100) = 30 [%]
If we would like the Trend Direction related parameter to have smaller impact on our trading position size coefficient, than we select Kt as 0 / 0.2 for example.
In this case the impact of this parameter would be:
((1+0.2) / 1 * 100 – 100) = 1.2 / 1 * 100 - 100 = 20 [%]
C: Market volatility – related parameter.
If we plan to have relatively smooth P/L curve on a long-term period, than this must be represented in our position size selection. So many traders go bust just because of this simple phenomenon. This parameter can be multiple things, like something, related to the Daily Average Range for the past N Days or a simple indicator, that is widely available, the VIX.
Obviously If the market volatility is high, than we need to decrease our trading position size accordingly so the position size – VIX relation is inverse.
The Average of the VIX during the past many years was between 20 and 25.
The Maximum VIX value during the last 10 years was above 80, the minimum was about 11 so the ratio between Max and Min is about 7.
This is way too much, but even during the past three years, the ratio of the VIX(Max) / VIX(Min) was about 4.
Yes we need to allow the VIX to have a relatively big impact on our position size selection, especially for those periods of extreme volatility, but still probable smaller, than the actual ratio of the VIX(Max) / VIX(Min)
If we want to transform the VIX impact to the “C” parameter in such a way, when the ratio between the C(Max) / C(Min) = 3, than we can select for example:
C(Min) = 1
C(Max) = 3
This will make it possible to alter the normal position size coefficient by 200 [%].
The reason to decrease the impact of the variability of VIX for the position size coefficient could be that over time and with some experience the brain and thinking process could adopt to the increased volatility slightly and intuitive traders learn to handle the increased risk.
But still we need this to restrain our trading during heightened risk in the market.
The following chart demonstrate the transformation of the VIX to the defined “C” parameter, and the equation below the chart describes the calculation of the Actual C parameter value, knowing the Actual VIX value.
D: Other Market-related or Technical indicator – related parameter.
This could be anything that we consider important for position size calculation.
It could be one specific technical indicator, like overbought / oversold condition, momentum indicator, strength indicator or a combination of indicators.
We can also include a Mental Strength Indicator if we are trading intraday, and we would like to have that represented.
It can be a continuous function, a step function or a set of data, that we define from other indicators, something like this: (Very Supportive, Supportive, Neutral, Against position, Strongly against position with corresponding data representation as of 5, 4, 3, 2, 1)
To clear up this a bit I present a few examples:
- The Daily prediction for the NASDAQ and the Omega prediction gives a Long signal.
- For this “D” parameter calculation we decide to use the CCI indicator (Commodity
Channel Index), call it TechParam, and the value of this will be the following, in support of a Long (Bullish) position:
TechParam = 5 If the CCI was below the -100 level during the past few days, but just turning upward Today and is about to cross the -100 level. (For CCI indicator users this is a Bullish situation.)
TechParam = 4 If the CCI was below -75.
TechParam = 3 If the CCI was between -75 and +75.
TechParam = 2 If the CCI is above +75 but the direction of the indicator is Upward.
TechParam = 1 If the CCI is above +75. but the direction of the indicator is Downward.
The D parameter, which can take five distinct value, can be viewed on the following chart.
Another example for a Technical indicator is that we use a channel (Which could be a trading range in a range market and a trend channel in a trending market.)
We split this channel into five equal size regions.
In the midst of a range market in the past few weeks let’s assume, that the predictor gives a Long signal.
Now in this case If the market index is in the top 20 [%] of the range, meaning that it is close to strong resistance levels from which the market turned back down, we enter a value for this TechParam = 1 as it is not supporting the Long position.
In case of a trending market it is more probable that the market participants will not interpret the top 20 [%] range as resistance, as the market moved higher recently in similar situations.
If the market is in the bottom 20 [%] area of a market range, than we enter 5 for this TechParam in a range market, as it is more probable, that the recent market behavior will prevail, and the market will find support at current levels and will turn upward with a higher probability.
Only our imagination limit the possibilities here.
All this might look a bit complicated, but it really does not.
To make the whole idea easier to understand I created a Demo Excel file and implemented “A”, “B”, “C” and “D” parameter there as potential impacts for our position sizing.
You can download this Excel file from HERE.
The name of the Excel file is Pos_Size_Calc_Tool_Demo and it can be downloaded from the predictionwizard download area:
http://www.predictionwizard.com/admin.php?p=downloads&act=downloads&parent=765
Please read the comments first, and after that you can experiment with it as you like.
It might be ideal to save the original file with a different name first and do your own experiment on a copy and not on the original file.
You can modify or tune it to your preference if you find it valuable for yourself.
It is a general purpose Demo Tool, that could be valuable for those, who consider position size and position management extremely important as I do.