Your Trading System Needs Internal Controls

I've said this before, but it's worth repeating - especially after reading about the rise and fall of Soul Trader's Grail forex system, which I always regarded as one of the most advanced automated systems around. Reading his explanation of why this system finally had to be shut down really brought home to me the need for rigorous internal controls on your trading system.

By internal controls I mean the type of tracking algorithms that watch over your trading performance and intervene to shut it down when performance begins to dip. Or to paraphrase Soul Trader, "How to identify when the market conditions cause the system to fail, and what to do when it's failed."

String of losses over X% of trading capital? Click: the system goes off until performance improves. Recent trading odds dip below Y%? Sorry, you're out of the market until things perk up - and if they don't perk up, you'll just have to be patient and appreciate the fact that at least your funds are safe. Preserving your capital in adverse conditions is pretty boring, but it's one of the most important single factors in trading success.

The only reason I'm still trading, and have even bothered to return to trading after a recent hiatus, is because of these built-in trading restrictions. I think of them as a firewall between me and the worst the market can dish out. They're not perfect, and they certainly don't prevent all drawdowns all the time, but they're a lot better than nothing at all. In fact, given the choice I'd much rather trade with no stop-losses and strong internal controls than vice versa.

Disclaimer: check back in a few weeks and see if I've abandoned my trading controls because they failed completely in unexpected market conditions.

Current Trading Strategy: Slow and Steady

There's a lot less daily drama and a lot more sitting on the sidelines in my forex trading these days, and so far it's a strategy that seems to be working. Because of the filters and meta-signals I've added to my trading system, it generates a lot fewer signals than it used to, since it's switched off during periods when it's likely to underperform. So in an average week I'll now make 2 trades or so, a lot fewer than the daily trades I used to make (and that ended up leading to some really nasty drawdowns late last year). I'm looking for smaller profits with each trade too, with my limit orders set at levels targeted for both trending and ranging price behavior.

This slow and steady strategy looks for a few high-probability trades and in the process filters out a lot of potential winners but also lots of losers, and seeks out modest profits across many different market conditions, with a focus on playing the likely daily range rather than hoping exclusively for profitable trends (which can be few and far between). And with the signal flipper I've added, there's a built-in ability to recognize when the core indicators aren't working and basically start fading my own signals.

Overall it's a slower, more patient, more boring and thus far, more successful approach to trading. So far this month I've placed three trades using this system and they were all winners, bringing in 31 pips in total profit. Again, not huge profits, but I'll take them over a drawdown anyday. The system also got the two trades before those right as well, making this a 5 trade winning streak. I'm under no illusion it'll last much longer, but I'm sure enjoying it while it does.

Signals and Meta-signals

If I had to categorize the different ways I've used trading signals in my brief forex career, I'd break them down into something like this:

1) I have no idea what a trading signal is, but I'm very eager to find out so I can get rich this year.

2) I'm trading with the first signals I read about in a book/website because someone said they work. But for some reason I'm not getting rich. In fact I've blown most of my account and if I'm going to preserve the rest I need a new style of trading - quickly.

3) I've discovered the psychological side of trading (the hard way) and am learning how to apply someone else's trading signals with discipline, rather than haphazardly whenever I need my trading fix.

4) I've begun to tweak the various off-the-shelf signals I've been using with my own variations and started using historical data and backtesting to determine which ones work best - if they work at all.

5) I've begun prospecting for my own trading signals using historical data and developing my own trading system that combines various signals with layered "and/or" logic. I end up focusing on this project obsessively for several months.

6) I discover (the hard way, as usual) that sometimes my trading system works well, and sometimes it's crap depending on underlying market conditions. To try and overcome this problem, I start exploring the idea of meta-signals, which monitor the performance of my trading signals and turn them on or off, or even reverse them, based on their recent track record.

So far the meta-signals I've developed act as trading filters, kill switches, and signal flippers. Others I've tried in the past but don't use currently are designed to weight competing signals based on their performance to determine which one is more likely to work. The basic principle is the same across the board: using internal feedback to fine-tune trading activity.

If you've been reading this blog much lately, you can probably tell #6 is the stage I'm in right now. Who knows, maybe it's the last signal-related stage before I start trading without any signals at all - though I can't imagine how I'd pull that off. I've found that meta-signals are as important as, if not more important, than the actual trading signals themselves. The reason is simple: they make my trading system work better. In fact sometimes I wonder if a really mediocre trading system could turn into a highly profitable one if the right meta-signals were attached to it.

So if you're ever frustrated with how your own trading system is performing, try running it through some meta-signals before trading it in for a new one. They might make all the difference.

The Kill Switch is Working

I haven't been doing a lot of trading the last week or so, which given current market conditions turned out to be a very good thing. I recently added a new set of filters to my trading signals and they create very strict conditions for when I can enter certain trades - and for the past couple weeks they've been switched on most of the time. I just went back and looked at how my system would've performed without this new "kill switch" in effect and it wasn't pretty: in total it's saved me from over 150 pips in losses in a little more than a week.

How does it work? Well, part of the equation is the Bollinger Band filtering I wrote about recently. The rest of the equation involves some internal feedback that would only really make sense to me if I described it so I'll spare you the details.

Another question well worth asking is why my system would have performed so badly in the current conditions. If you've been watching the EUR/USD for the past several days you'll see why immediately: lots of very range-bound activity, with false breakouts and whipsaws that all seem to settle back into the same general price zone day after day. This is the type of market that would absolutely destroy my trading account if I let it - but with the new trade filters in place, that's less likely to happen. (I hope.)

Related topics:
The Quest for a "Kill Switch" Signal
Bollinger Band Trade Filtering
Creating Self-Tuning Signals

Bollinger Band Trade Filtering

In the spirit of my New Year's resolution to put my trading system on a diet, I've been exploring ways to more aggressively filter out unnecessary trades, boost my trading odds, lower spread costs, and generally fine-tune my system to optimize its performance.

One filtering method I've found that shows a lot of promise involves Bollinger Bands, but not in the way they're usually used. Until now I used Bollingers to identify points where the price was likely to begin a reversal from a (relatively) overbought or oversold position (in forex, all things are relative). But for my current trade filtering project, I'm using them to identify periods of low volatility, when the EUR/USD pair is trading within a narrower range.

The measurement I'm using to determine these periods is the distance between the upper and lower Bollinger Band. When the bands are further apart, the market tends to be in a more volatile, wider-ranging phase. My forex strategy works best in these types of periods (as do most trading systems, I suspect). When the bands are close together, volatility is lower and my system tends to accumulate losses and trading costs.

So what I've done is have my trades switched off when the Bollinger Bands are too close together; the optimal distance between them is something I've determined through a lot of backtesting. What I've found is that my trading odds and performance improve significantly, with fewer big drawdowns and pointless trades when the markets are in the doldrums. Of course, because Bollingers are a lagging indicator, I also miss some big breakouts when the market shifts back from low to high volatility. But I'm willing to live with that - consistent trading performance is far preferable in the long run, and there's nothing that can kills your morale more quickly than waiting out a series of bad trades when the market's drifting in no particular direction. The more of those ugly periods I can avoid, the better!

Related topics:
A Nice Bollinger Band Trade
Put Your Trading System on a Diet
The Quest for a "Kill Switch" Signal

Sometimes No Signals are Good Signals

Today was the best example yet of when following my trading system to the decimal point pays off. Yesterday I plugged in all the relevant data to generate today's trading signal. And my system, somewhat to my frustration, didn't generate either a short or long signal. As I always do when I don't get a signal I considered ways of forcing one, which is how my urge to overtrade currently expresses itself. But I decided it would be foolish to end a profitable week by breaking my own trading rules just to get in one last trade.

That rule-breaking trade would have been a long one, since that's what forcing a signal would've given me (I've done this before, can you tell?) This morning when I checked in on the EUR/USD out of curiosity I saw the results of today's Non-Farm Payroll announcement: a decline of around 100 pips. If I'd placed my undisciplined trade I would've started the weekend on an extremely bad note, having broken my trading rules and paid a high price as a result.

What I find intriguing is that my system avoided trading today, not because I've coded it to avoid high-volatility news events like the NFP, but based entirely on technical price analysis. I can't point to any one equation that made this decision for me, but I have to wonder if there was a specific price movement that signalled a period of dangerous volatility on the horizon. This mysterious region where technical analysis intersects with real-world news is fascinating, because I feel technical indicators can sometimes identify a safe or hazardous news trade before it occurs.

Or maybe it's my imagination - what do you think?

The Quest for a "Kill Switch" Signal

One of the things on my forex wish-list is an anti-signal that effectively blocks trades at difficult points in the market. I think of this as a "kill switch" that would turn off my entire trading system at times when the market's behaving so strangely that it's too dangerous to be trading. For example, my system tends to fare worst in situations where price is ranging through a fairly narrow channel and no clear long or short trend has emerged. In ranging markets, my signals keep looking for that next step up or down in price that would logically follow in a trending market, but which in a ranging market often turns into an abrupt reversal instead. So I'm always staggering along one step behind, and losing money on each reversal. Here's a good example of a ranging EUR/USD market that cost me a lot of money in May and early June of this year:

Essentially, then, I'm looking for an accurate way to detect ranging market behavior early on so I can disable my trend-based signals. A nice bonus, once I've gotten that switch built, would be to develop a whole new set of signals for ranging markets that would be activated by the same switch. (After all, as I noted in my process of elimination post, turning off one signal often means you're turning on another one.)

In some ways this may be a fool's quest, based on an impossible ideal: namely, that you can easily identify a ranging market and start making accurate trading decisions before it becomes a trending market again. So far all of the indicators I've tested have too much of a lag built in, and just as they come to the conclusion that the price is truly ranging, it heads off on an upward or downward trend. For instance, in the illustration above, the market suddenly began a major downtrend and sank around 400 pips by mid-June. If I had a kill switch operating, would I have been able to get my trending signals turned back on in time to turn a profit? I'm coming to the conclusion that one of the costs of having sensitive trending signals is having a high failure rate in ranging markets. As long as the market is trending more than it's ranging, you'll do OK, but you'll be in for some painful stretches along the way.

But enough theorizing. The fact is, I already have all kinds of kill switches built into my system that I'm just taking for granted. For instance, a stop-loss is essentially a kill switch. So are my signals that prevent any short trades when certain long signals are generated, and vice versa. So are situations where neither a long or short signal is generated.

I'd still love to find a quick way to detect ranging markets, though - so if you have one you're willing to share, please post it in the comments!

Related topic: Trading by the Process of Elimination

Creating Self-Tuning Signals: How Internal Feedback Can Improve Your Trading System

While trying to improve the performance of a trading system I was backtesting, I found myself wishing there was a way it could turn itself off, or at least improve its odds, when it was performing badly. If for some reason the current market dynamics didn't agree with my system, why not allow the system to recognize the adverse conditions and take a hiatus until the trends returned to behavior it could make sense of?

On other occasions, I wished for a feature that would track particularly good signal performance, and would trigger additional trades outside my primary signals because their trades were going so well in a particular direction.

Essentially, I was looking for ways to amplify good performances and cut short bad ones. So I created what I call monitoring signals or meta-signals, which keep tabs on how the primary trading signals are functioning and have the ability to turn them off, to trigger a trade in the opposing direction, or to trigger "bonus" trades in the same direction if the signals are performing optimally. Here's how they work:

One monitoring signal observes the performance of long trades generated by my forex system. If the past 5 days of trading have generated over 150 pips in profit, this feedback circuit automatically triggers another long trade on the sixth day. In my backtests I've found this meta-signal provides a significant boost to system profitability.

Another signal compares the performance of my long and short signals. It counts how many long vs. short signals have been generated on the past 5 trading days. If there have been more long signals than short ones, and if on the current day no long signal has been generated, it automatically triggers a short trade. The idea being that after a run of long signals in the market, it's statistically likely to rebalance itself with a downtrend. On the short side, I have another signal that performs exactly the same function, stepping in to trigger long trades after a run of short ones.

Some other types of meta-signals I've experimented with, albeit with variable results, have measured the slope of trading gains on a chart, and have excluded trades made when the slope of recent gains is either too gradual, or is negative. Another similar feedback signal measures the percentage of successful trades generated by a particular signal, and excludes trades made when its success over the past N trades has been under a certain percentage.

I could go on, but I'm sure you get the idea by now, and most of my other meta-signals have been variations on these basic themes. When designing a trading system, it's always good to keep in mind that in addition to having it analyze the forex market, you can also have it analyze itself.

Trading by the Process of Elimination

When I started creating and testing long and short trading signals, I discovered a whole class of signals that hadn't occurred to me previously. They're trading indicators generated by a process of exclusion or elimination: if long signals A, B, and C are NOT present, then by default a short signal is generated. What may not be immediately obvious when looking at a forex chart is that the absence of certain indicators can be just as important as their presence.

In some cases, the absence of some signals can amplify an opposing signal -- for example, here's the syntax of one Excel formula I've created that triggers a signal only if one long indicator is present while three short indicators are absent: =IF(AND(W572=1,Z572=0,Y572=0,K572=0), where 1 is a long signal and 0 is the absence of a short signal.

I could give a bunch of other examples, but you probably get the idea. So if you ever set out to design your own trading system, remember that you when you create a long signal, its absence may also be a short signal, and vice versa. Who knows how many pips this could add to your trades!