Fisher Transform - a sample size problem?

March 11th, 2011, 05:01 AM

wccktrader

I have backtested it for US equities. As the indicator is based on price distribution (pdf), it does a good job in finding cyclical turning points. It is also less sensitive to the lookback period as compared to other momentum indicators such as RSI, CCI and Stochastic. Such indicator would work well if their lookback periods are in-sync with actual cycle period of the market. However, they often give poor results when the actual cycle period of the market changes. I suppose this is one of the reasons for their poor walk forward testing results.

Wccktrader,

Thanks very much for the reply and thoughts.

Many I ask what your results were with the US Equities? Again, I�ve done some rough and ready testing, but thing robust as yet. I�m hoping to progress on that front today. My initial results show that it really needs to be traded on the right timeframe with a portfolio of equities at the same to smooth the equity curve and the right timeframe is important.

Thanks for sharing your views on the cycles. Would it be ok if we kicked that idea around a little? I thought that Ehlers had achieved with applying the Fisher Transform to the channel of prices is to create a Gaussian like distribution, therefore, the cycle is less relevant as the PDF is build regardless of the cycle and should be basically Gaussian always. Have I misunderstood? Ehlers focuses on cycles with his other work on the MAMA (https://www.mesasoftware.com/Papers/MAMA.pdf). Where he looks at the application Sine waves to identify the cycle.

Thanks and regards,

drolles

March 11th, 2011, 05:11 AM

Trader.Jon

Do you have a specialized workbook for doing that directly out of NT? I have a feeling that is only a small part of what you have done in terms of analysis.

Jon

Jon,

No, no specialised workbook. You can achieve the same thing by exporting the optimisation results using the usual method of right clicking on the Strategy Analyser Optimisation results tab, exporting to Excel then running a Pivot table with the performance parameters. Given the way that NT exports to the spreadsheet, you usually have to use Text to columns to break the parameters into separate columns.

Kind regards,

drolles

March 11th, 2011, 12:00 PM

drolles

Wccktrader,

Thanks very much for the reply and thoughts.

Many I ask what your results were with the US Equities? Again, I�ve done some rough and ready testing, but thing robust as yet. I�m hoping to progress on that front today. My initial results show that it really needs to be traded on the right timeframe with a portfolio of equities at the same to smooth the equity curve and the right timeframe is important.

Thanks for sharing your views on the cycles. Would it be ok if we kicked that idea around a little? I thought that Ehlers had achieved with applying the Fisher Transform to the channel of prices is to create a Gaussian like distribution, therefore, the cycle is less relevant as the PDF is build regardless of the cycle and should be basically Gaussian always. Have I misunderstood? Ehlers focuses on cycles with his other work on the MAMA (https://www.mesasoftware.com/Papers/MAMA.pdf). Where he looks at the application Sine waves to identify the cycle.

Thanks and regards,

drolles

I would backtest on a basket of stocks and choose the stocks that show good results for trading. These stocks are usually stocks that are trading in upward or downward sloping channels i.e. stocks trading in smaller cycles within upward/downward sloping channels.

I think the Fisher Transform has been designed with the purpose of clearly identifying cyclical turning points. This is given in the conclusion of the paper by Ehlers attached in your earlier post -

"Prices do not have a Gaussian PDF. By normalizing prices or creating a normalized
indicator such as the RSI or Stochastic, and applying the Fisher Transform, a nearly
Gaussian PDF can be created. Such a transformed output creates the peak swings as
relatively rare events. The sharp turning points of these peak swings clearly and
unambiguously identify price reversals in a timely manner. As a result, superior
discretionary trading can be expected and higher performing mechanical trading
systems can be developed by using the Fisher Transform."

March 11th, 2011, 05:21 PM

As per the update previous, I�ve adjusted an existing script a ran the strategy across the Dow 30. It threw up some interesting results. The first one to note is the low Profit Factor at 1.05. However, again for out of the box not too bad. However, a couple of things that will need some attention. 1. The variance of the pay offs. I�ve attached a graph of the daily PnL. 2. The number of trades on some days. I suspect it is opening a number of trades at the same time and probably exceeding any sort of risk measures (e.g. max 5% of account at any one time). I�ve attached a of the number of trades per day vs the payoff for that day.

I think like my post previous on the Meander strategy, I think we are looking at a system with many of the same characteristics.

The other thing that I�m thinking is that it definitely could be used across a bigger set of instruments, like the whole S&P500. However, if definitely needs to be wrapped up with some sort more robust risk management before it is progressed.

The cumulative profit curve is the backtest results using a 50,000 account with a Fixed Fractional allocation of 1% per trade from 01/01/2008.

Kind regards,

drolles

March 13th, 2011, 08:27 AM

I�ve now run a combined test on the S&P 500. Unfortunately, not good results. I�ve not constructed this as a portfolio of the whole S&P500. I�ve done it as a single script then run it across the individual instruments. As you can the results are not worth taking any further. We are looking at an almost breakeven with a very large number of trades (nearly 40,000) across a year.

March 14th, 2011, 01:12 AM

drolles

I�ve now run a combined test on the S&P 500. Unfortunately, not good results. I�ve not constructed this as a portfolio of the whole S&P500. I�ve done it as a single script then run it across the individual instruments. As you can the results are not worth taking any further. We are looking at an almost breakeven with a very large number of trades (nearly 40,000) across a year.

I have never found it to be easy ... each instrument is like its own universe and needs to be massaged to have the best probable outcome.

I highly doubt that there are more than two out of that list of 500 that have comparable/similar results?!

Jon

March 15th, 2011, 07:06 AM

Jon,

Yes, you are right. The results were highly variable within the S&P500. Very big differences between the performance each of the stocks within that. However, given the high correlation of the instruments within the collection, the bad days and the volatility were all aligned. I�ve been thinking is there is there anything one could do at the portfolio level to manage that risk and correlation (returns and volatility).

I�ve been doing some testing with portfolio measures of drawdown which appears to improve the performance (note here, not overall cash returns, but improve Sharpe and Profit Factor).

Kind regards,

drolles

Fisher Transform - a sample size problem?

Discussion in NinjaTrader

Fisher Transform - a sample size problem?