Van Tharp's Position Sizing Game

[Gringotts Bank]

I've just downloaded and had a play with this. First 3 levels are free.

I don't get it! What can be learnt from this other than don't go over 2% and don't go too much under 1%? Am I missing something?

Why does he make it so that you have to sit there with your calculator entering the same thing over and over? Wouldn't it be better just to type in the % risked for every trade and then hit "backtest" and view the equity curve... and then compare them all for different sizes?

 

[howardbandy]

If I may offer an opinion ---

Position sizing depends on the synchronization between the model (logic, rules, and parameters) and the data being processed and traded. It is not possible for the model to determine what the position size should be. That should be done by an analysis of the distribution of the trades. Having Any position sizing, including compounding, in a trading system creates a bias based on the specific sequence of trades encountered during development. That bias overestimates profit and underestimates risk.

While it is interesting and potentially helpful to learn about betting systems, such as described by the Kelly formula, those betting / position sizing techniques are applicable to processes / games / systems that have much more stable characteristics than trading. They work best when the distribution of bet returns is stationary (as a roulette wheel) and the amount won equals the amount lost -- neither of which is true for trading results.

Be wary when applying simplistic position sizing. It is important to analyze the entire distribution of returns -- not just the mean, or even the mean and standard deviation. The critical limitation to maximum position size is the number and magnitude of losing trades, but the entire distribution must be taken into account.

Evaluation of risk is very subjective, but it can be quantified. Every trader should quantify his or her own personal risk tolerance. A sample risk tolerance statement is:
"I am willing to accept a 10% risk that the maximum intra-trade drawdown (measured from highest equity to date) over the next two years will be no greater than 20%."
The analysis techniques I describe in "Modeling Trading System Performance" estimates the maximum position size for the combination of:
1. A specific set of trade results
and
2. A specific statement of risk tolerance.

As real (or paper) trades are made, those trade results should be used to keep the "best estimate" set of trades up to date in a Bayesian-like manner. This gives the trader the tools needed to monitor system health and adjust position size as system performance varies.

Best regards,
Howard

 

[Gringotts Bank]

If I may offer an opinion ---

Position sizing depends on the synchronization between the model (logic, rules, and parameters) and the data being processed and traded. It is not possible for the model to determine what the position size should be. That should be done by an analysis of the distribution of the trades. Having Any position sizing, including compounding, in a trading system creates a bias based on the specific sequence of trades encountered during development. That bias overestimates profit and underestimates risk.

While it is interesting and potentially helpful to learn about betting systems, such as described by the Kelly formula, those betting / position sizing techniques are applicable to processes / games / systems that have much more stable characteristics than trading. They work best when the distribution of bet returns is stationary (as a roulette wheel) and the amount won equals the amount lost -- neither of which is true for trading results.

Be wary when applying simplistic position sizing. It is important to analyze the entire distribution of returns -- not just the mean, or even the mean and standard deviation. The critical limitation to maximum position size is the number and magnitude of losing trades, but the entire distribution must be taken into account.

Evaluation of risk is very subjective, but it can be quantified. Every trader should quantify his or her own personal risk tolerance. A sample risk tolerance statement is:
"I am willing to accept a 10% risk that the maximum intra-trade drawdown (measured from highest equity to date) over the next two years will be no greater than 20%."
The analysis techniques I describe in "Modeling Trading System Performance" estimates the maximum position size for the combination of:
1. A specific set of trade results
and
2. A specific statement of risk tolerance.

As real (or paper) trades are made, those trade results should be used to keep the "best estimate" set of trades up to date in a Bayesian-like manner. This gives the trader the tools needed to monitor system health and adjust position size as system performance varies.

Best regards,
Howard

For a discretionary trader, would you suggest something like Market System Analyser to fine tune position sizing? If so, what number of trades would you say makes up a reasonable sample for this process.

Also can you explain this please? "As real (or paper) trades are made, those trade results should be used to keep the "best estimate" set of trades up to date in a Bayesian-like manner".

Thanks Howard.

 

[tech/a]

Good question GB
I would also like to see Howards take
On a discretionary traders options.

 

[Gringotts Bank]

Good question GB
I would also like to see Howards take
On a discretionary traders options.

This looks interesting (Tharp's SQN), though not a lot of info given. http://www.vantharp.com/tharp-concepts/sqn.asp

 

[Boggo]

For a discretionary trader, would you suggest something like Market System Analyser to fine tune position sizing? If so, what number of trades would you say makes up a reasonable sample for this process.

Good question GB
I would also like to see Howards take
On a discretionary traders options.

A pictorial view of the considerations.

Below are two examples with the same entry, same target and same $$ at risk in both, the only difference from a discretionary perspective is where you would place the stop.

The question to be tested I suppose is how often you would get stopped out in example 1 vs example 2 and is it worth even attempting a trade in example 1 considering the reduced R/R.
(click to expand)

 

[howardbandy]

Greetings --

Systematic analysis of risk, position sizing, and profit potential requires unambiguous identification of trades. This is fairly easy with formula-based systems, and more difficult with discretionary systems.

There is little doubt that developers of formula-based system introduce a number of biases into their results -- all of which result in an underestimate of risk and an overestimate of profit potential. But the difficulty we all have identifying patterns subjectively is a more serious limitation. [Two excellent books that helped me understand my own limitations with observation and intuition are Daniel Kahneman's "Thinking, Fast and Slow," and "The Invisible Gorilla" by Christopher Chabris and Daniel Simons.]

Whatever the source of the trades -- real or hypothetical, discretionary or formula-based -- the analysis of risk begins with a distribution of trade results. A trade list works well, or a probability distribution. That set of trades is what I call the "best estimate" of future performance. For systems that have gone through walk forward validation, it is the out-of-sample trades from the walk forward runs. However that set of trades is created, it should be the developer's best estimate of future performance. If desired, you can add trades that you think might occur in the future that are not already present in the best estimate set. Or you can create and use a hypothetical set of trades -- a technique that is useful for deciding whether a particular system is likely to be tradable before expending too much energy fully developing it.

There is an outline of the development and trade management process I recommend in Chapter 2 of my book "Mean Reversion Trading Systems." It is free and can be downloaded from the book's website:
http://www.meanreversiontradingsystems.com/book.html

Best regards,
Howard

 

[howardbandy]

This looks interesting (Tharp's SQN), though not a lot of info given. http://www.vantharp.com/tharp-concepts/sqn.asp

Greetings --

Search the ASF threads for my postings over the past five years or so related to the use of Dr. Tharp's System Quality Number.

SQN is based on the t-statistic, but Dr. Tharp has modified the calculation so that the t tables cannot be used with confidence.

Let me reiterate two important points --

1. The rules that identify patterns -- whether these are formula-based or subjective -- cannot determine the health of a trading system, so cannot determine safe position size. There should be no position sizing calculations, not even compounding, during the development of a trading system. Including any position sizing introduces a bias that results in an overestimate of profit and an underestimate of risk.

2. The performance of a trading system varies over time as the data goes through phases where the patterns and signals are easily and clearly identified, where they are difficult to identify, and where there are many false signals. The performance of all trading systems is not stationary. The maximum safe position size changes as the data changes, and must be recalculated regularly. All systems eventually degrade. As a trader, you should be reducing position size as performance degrades. Eventually, as the system fails completely, the correct position size for a system that is broken is zero.

--------------

Be very careful when considering implementing position sizing using the techniques Dr. Tharp recommends.

Also, readers of his book "The Ultimate Guide to Position Sizing" should realize that many of the examples of trading system results in that book are completely unrealistic. A trader with a reasonable amount of experience would be able to retire comfortably after a few years of trading a system that trades frequently and has a t-statistic of expectation of 3.0. Recall that a t-statistic of 3.0 means the results are three standard deviations greater than the mean. Even thinking about systems with t-statistics of 7 or 10 in truly out-of-sample results is beyond fantasy. [Typical trading systems have a standard deviation 3 or 4 times the mean. If your system has a standard deviation less than 2 times the mean, it is outstanding.]

Best regards,
Howard