I want to reiterate for the record my take on one detail of bot design that came up in the discussion with my son last weekend, to do with why IMO a bot will only ever trade as well as its creator but never better, and more often than not considerably less well as many are quick to point out.

Recent experience adding S/R energy to a bot styled after Barry Burns' Top Dog trading system, which has been under construction for almost 2 years now, brought home the distinction between tweaking a bot and improving it. As others on nexusfi.com (formerly BMT) have noted, early in the bot craft learning curve when comparing the price chart with bot trading performance we may be inspired to modify the bot to handle particular instances of price movement (missed profit opportunities or "unnecessary" losses), switching perspective from the general to the specific (from the system to those devilish price details), an approach commonly called "adding filters" or the like.

This initial propensity to tweak bots for the wrong reasons may stem from our earliest impressions of automated trading, the evolution of which for some of us might be summarized as follows: having no trouble picking highs and lows on the first charts we encountered we conclude we are natural born traders & start planning retirement accordingly. Subsequently bloodied but not bowed after doing battle at the brutal right edge of the chart for a while, we surmise that a bot could be the antidote to new & unexpected feelings of confusion, uncertainty & self loathing because a bot does not feel. Surely bot writing requires not much more effort than we've put into manual trading since all it has to do is "look back 50 bars and do what it should have done back then", which is how WE learned to trade, after all, and furthermore a bot can buy the lows and sell at the highs 24/5, freeing us in the meantime to get a head start on the lounging around and umbrella drinks while the bot toils unattended to build our fortune. Still later, after many more blind alleys and Holy Quests, we conclude that if we can't trade we can't teach a bot to trade either, sadder & broker but perhaps more appreciative of the sarcasm in the remark, "Those who can do; those who can't, teach". Which pretty much explains the Interwebz.

Arbitrarily "adding filters" may be educational when the bot is being used to test ideas but if being developed to implement a potentially profitable trading system this approach--switching focus from general behaviour to performance artifacts--is doomed from the start.

Assuming we capture the essence of a system in the first instance bot performance reflects 2 things:

1. the degree of fidelity of the implementation; in other words the extent to which the bot embodies what is necessary to the system; and,
2. the sufficiency of the system being coded

Any performance issues have to do with one of those 2 factors, and not ever with particulars of price behaviour.

Bot performance can be improved only by incorporating additional aspects of the thought process when trading the system that might have been excluded in the first cut (e.g., due to project management issues).

Except for looking AT the chart as a debugging aid therefore a rule of thumb when developing a bot to implement a trading system may be to focus only on the performance report--hence to avoid altogether looking TO the chart in the hope of improving performance. If testing shows any kind of perspective switch is warranted it must be toward greater abstraction (i.e., in terms of a different system).

The above remarks do not address the fact that modifying a bot to deal with chart particulars moreover defeats the purpose & effect of backtesting (a point made by Kevin Davey in his webinar), merely curve fits the bot to the test set, which is in the category of engineering uselessness.

A daughter who is doing a graduate degree in neuroscience at Western University (Ontario) in town for the holidays still dismisses any suggestion I make to study traders' brains ("probably full of holes" is all she will offer) but did say she had the opportunity recently to chat up Terry Stewart over dinner when he delivered a lecture on Spaun (the "big artificial brain" that made the news a while ago. Terry Stewart is a Post Doc at U. Waterloo Centre for Theoretical Neuroscience and a Spaun team member). She couldn't answer all of my questions about the model so I downloaded the paper from the distribution site here. A quick read renews my interest in that sort of research but the model doesn't look like anything one could implement any time soon for trading.

Just a quick observation waiting for IB to finish rebooting its servers.

Basic indicators like MACD, stochastics and SMA's fascinate me so still in recreation mode earlier tonight I reviewed the 2nd installment (continuation and advancement) of [MENTION=2891]perryg[/MENTION]'s method (webinar here) and [MENTION=11447]NJAMC[/MENTION]'s work to automate an earlier version of perryg's system (automation thread here)

The reason for doing so stems primarily from the hypothesis that a successful system has elements in common with every other successful system even if terminology may vary and these elements may be expressed somewhat differently. By virtue of these difference perspectives therefore comparing systems may be one way to increase depth of understanding of a given system.

A number of differences in perspective between Barry Burn's & perryg's systems IMO include the following:

1. what perryg calls "divergence" (measure of spread between moving averages) appears to be BB labels "momentum";
2. "fractals" (definition of which in perryg's system attributed to Bill Williams in "Trading Chaos") appear to be what BB calls (& essentially dismisses as) price pivots.
3. inclusion of indicators from higher time frames what Barry Burns refers to as fractal energy, both serving as entry confirmation
4. in any event neither of these definitions of "fractal" matches my definition of it, which is more along the lines of the Hurst indicator described in previous posts.

What both appear to agree upon, and IMO is the alpha and omega of Ichimoku, is the the significance of moving averages. In my own experience simple moving averages are a species of low pass filter (not a very good one) with a lag equal to the length divided by 2 and a messy impulse response as follows (courtesy of course notes for an engineering course at Berkley,

H(w) = (1/L) * ( 1 - 1 / exp(jwL)) / (1 - 1 / exp(jw)).

where H = amplitude, w = angular frequency, L = MA length, j = sqrt(-1), exp = exponential function (base of natural logarithm "e" to the power of...)

Figure 1. Normalized magnitude (amplitude) for 3 MA lengths: L = 4 (red), 8 (green), and 16 (blue) (from the same source as the equation)


Abrupt changes in an MA due to early samples dropping out aside, in the figure the points at which the magnitude of the response falls to zero signify that price energy of corresponding wavelength is suppressed in the current value of the SMA, whereas price energy of wavelength corresponding to response peaks is expressed in the current value of the SMA. While the existence and cause of cyclical price energy may be debatable (i.e., "event Y occurs on the 20th day of every 3rd month"), AFAIK it is what was originally the basis for choice of MA length, including Tenkan-sen and Kijun-sen lines of the Ichimoku indicator, possibly predating e.g. the sentiment that "everybody is using an XXX MA and that is what explains their power as S/R levels".

The art & science (magic & mystery?) of fast and slow SMA crossovers is another topic entirely, but IB is back up so back on our heads.

Not strictly relevant to this journal (which has to do with spot Forex system development) spent most of today watching webinars by [MENTION=25132]FuturesTrader71[/MENTION], some from his site and all 4 nexusfi.com (formerly BMT) presentations listed in this post and reading posts in volume profile threads in preparation for tackling a volume profile & footprint trading plan.

One thing this exercise has taught me so far is it really is no contest in terms of data quality & responsiveness between IQFeed and IB as data sources

Scalping small sizes and single targets in several spot currencies on the cash account through the European open this morning to see the effect of some changes to the spot trading plan, one of which was inspired by a remark [MENTION=25132]FuturesTrader71[/MENTION] repeated often in his videos; namely once the trade is entered don't mess with it.

A second change is to forego momentum exits in favour of strict S/R exits and the combination of changes is resulting in trades like the 2 illustrated in the charts below. Previously (as a finger-on-the-button scalper) I would very likely have been out with 1-2 pips the instant momentum faltered, so this is something new and painful for me. Setups and entries are still determined according to BB's system, and I attribute the fact the small number of trades I've made thus far have been successful to the system. In any case I don't mind the boost to "emotional capital" even if it's short lived.

Unfortunately it's early days and until I can code BB's setups for backtesting the effect of any change to the plan must be determined by manual trading, which means 100's more trades before the potential value can be estimated.

Once MC is configured for volume profile and cumulative delta the plan is to see if those can be incorporated into the system (using futures to cue spot trades). I suspect this approach will not work for the sort of time frames (ultra short) we've dealt with so far, so that work may be the preamble to trading longer time frames.

Figure 1: short AUD.USD


Figure 2: long EUR.USD

Having spent most of today working with MC 8.5 beta volume profile and cumulative delta implementation I'm starting to miss the good old days when if a feature needs tweaking all one had to do was modify the open source NT code on nexusfi.com (formerly BMT). IMO MC's cumulative delta works well enough but volume profile features need work.

As a consequence I've dusted off NT, done a quick sanity test with IQFeed and installed [MENTION=1469]gomi[/MENTION]'s code per [MENTION=8812]redratsal[/MENTION]'s excellent video and started reviewing the many threads on the topic.

This venture is still in the category of recreation and in no way impacts the bread and butter trading plan, merely revisits topics that sent me into info overload when I first joined nexusfi.com (formerly BMT). Learning how to trade is an iterative process after all, and I'm only a few years into it. On that note, having stumbled across [MENTION=544]Zondor[/MENTION]'s contributions at every turn would take my hat off if I wore one. Instead I'll fuss with my do-rag (keeps the hair out of my face).

For anyone who still follows me, my wife accuses me of autistic tendencies (and she should know as a professional in the field) but please bear in mind I carry along everything I've ever thought about looking to apply it.

Oh look, there goes a butterfly.

Edited to add. If I'm autistic then the hallmark of autism is the autistic are simply bored beyond imagination.

Does not affect IQ and does not affect the ability to know when fundamentals indicate a particular economy, government, stock or currency is FUBARed.

Yep. Lately I like to short things.

bnichols 

For anyone who still follows me, my wife accuses me of autistic tendencies (and she should know as a professional in the field)

Yes I'm following and really enjoy your thread. We seen to share a lot, both in our trading lives and personal lives.
Personally, I have a mild form of " Asperger's Syndrome. ". My father is a full Aspi. Have your wife have a look at this and see how you fit.

Regards,
Neil.

Oh boy back to NT hallelujah!

I've been really enjoying your TDStochastic, it's helped my trade entry quite a bit.

Just a quick note mostly as a reminder to myself, taking another look at time of day stats decided to rework the (MC) indicator to process future volumes from IQ data the following preliminary unannotated plot (Open Office Calc) of raw data (unfiltered e.g. for dead time on Fridays and Sundays) made me do a double take.

The plot shows cumulative hourly volume (Y axis) vs EST hour of the trading day (X axis) for the last 120 days from a 6e-03-13 200 tick chart. Guess we know when retailers with day jobs tend to trade