Where's the math, and what I've been learning

Beginner Questions
1

I’m wondering where I can find hard data on the actual probabilities behind trading strategies. I see a ton of qualitative judgment calls (good, bad, super, etc) and very little in the way of numbers.

For example - Bollinger bands. How often do they successfully describe trends? 50%, 60%, 40%? Does the accuracy change dependent on other factors? (That’s a rhetorical question - I’m looking for a central resource that covers this, not a direct answer.)

I want to find objective analyses of technical indicators, and I don’t want subjective assessments. Does anyone know where I can find the hard numbers?

Wikipedia covers a lot, and has good references, but isn’t directed at Forex traders.

What I’m seeing and starting to believe is that a majority of market behavior looks like a random walk. Without Fundamental knowledge of some sort - whether implicitly encoded in the strategy or explicitly using real analysis - a strategy is naive to the drivers of the market.

An implicit strategy would incorporate an assumption about the market - a Fundamental judgment call embedded in a technical indicator. Something like “in the morning in the US, markets are likely to be perky, therefore I can capitalize on trends of type X.” Moving averages and crossovers of particular scopes could be used to signal such implicit assumptions.

I’ve noticed that market activity holds certain patterns year to year - each month, season, holiday, major sporting event, and other significant annual human event has an impact on markets. This leads me to conclude that there is profit to be made in knowledge about these changes. Any event that significantly impacts human events will significantly impact currency trading in discrete ways - it will change volatility or relative value.

Human life ultimately drives currency exchange. Knowledge about perception of value, real value, and behavioral patterns are the only rational way to gauge market trends. However - technical analysis can incorporate that knowledge implicitly.

I’m beginning to think that I’ll need to become something of a historian and psychologist in order to understand how the markets work. At any rate - does anyone know where I can find quantitative, objective analyses on Forex trading strategies, or is this pretty much what every trader has to learn for themselves?

2

"I’ve noticed that market activity holds certain patterns year to year "

I’ve noticed that price action has certain patterns that happen ‘almost’ every day, every week, every month. No numeric analysis but I would guestimate 90% of the time. Maybe I should analyze the numbers on this just to get something more concrete than just a guestimate.

3

Don’t think you’ll find anything like that. Traders tend to play it close to the vest with that sort of information - those who’ve actually done that sort of research in the first place, that is.

What I’m seeing and starting to believe is that a majority of market behavior looks like a random walk.

Keep in mind that “looks like” is not the same as “is”.

An implicit strategy would incorporate an assumption about the market - a Fundamental judgment call embedded in a technical indicator. Something like “in the morning in the US, markets are likely to be perky, therefore I can capitalize on trends of type X.” Moving averages and crossovers of particular scopes could be used to signal such implicit assumptions.

To the extent that price incorporates the expressed views of market participants where the fundamentals are concerned, anything derived from price also does so.

I’m beginning to think that I’ll need to become something of a historian and psychologist in order to understand how the markets work.

If you’ve intellectually inclined, look into the subject of Behavioral Finance.

4

Thanks - definitely moving toward what I want to learn. I’ve been making trades, but I’ve also been looking at algorithms that can help me learn more.

One thing I’ve done is download the last 11 years of data for 18 pairs, and what I’d like to do is identify and correlate major events in the market with events in the world. Wikipedia has a nice resource in their “dates in history” - I’ve already started parsing the data into a sqlite database, and then I’ll be able to identify and classify events using simple scripts. When I can get a record of events, I can then use Wikipedia (or Nexis, eventually, or some other news repository) and start learning how to use news in trading.

One thing I’ve been wondering is after I’ve got my database nicely cleaned up, would it be worth anything to other traders? That would be a nice way of kickstarting my stake. The advantage of SQLite is that its both portable and structured - you can use almost any scripting language in existence with standardized queries to analyze data.

I’ve been trading randomly on my OANDA demo account, figuring out how to set stops and take profits and all that jazz. Really hoping that this will click.

My observation about random walks is based on the idea that over time, in a random walk, any strategy will encounter periods of profit and of loss. A good money management strategy could successfully keep you afloat in a random walk scenario for a really long time, even though the actual buy/sell strategy has no functional relationship to the trends in a market. This means that in a system displaying all the characteristics of a random walk, arbitrary, disjunctive strategies can make money some of the time(even though we know that the generative function is not random.)

I want to learn how to separate the random from the non-random. It’s not enough for me to see a strategy, and see evidence of its success, without understanding why - because if there are no solid reasons for success, there’s a strong likelihood that the strategy has no relation to reality, and is just encountering positive variance.

Once I’m able to dismantle the randomness into a picture of moving forces, I should be better prepared to trade live. I don’t necessarily want The Prime Radiant, but a reasonably accurate model with measurable variance should be possible.

5

Backtesting can be difficult especially if you cannot account for live spreads and live execution. How automated is your system? Is everything automated including entries and take profits/trailing stops?

Or is some of it discretionary?

General hard numbers can never be found, because everyone will trade the pattern differently. Some people trade bollinger bands with different stops and exit strategies than everyone else. There are just so many combinations.

6

I’m only trading manually. I like to be able to understand what I’m doing before I get anywhere near automating it, since it lets me avoid a whole lot of bugs. I’m doing more experimenting with [I]how[/I] things work than [I]why[/I] at this point.

General hard numbers can never be found, because everyone will trade the pattern differently. Some people trade bollinger bands with different stops and exit strategies than everyone else. There are just so many combinations.

That’s why the markets display the characteristics of a random walk - there are so many influences that play a part in price changes that its difficult to separate them out. However, we know there are influences to be found that change the random walk - I’m talking about being able to model those influences. In a very general sense, those influences are the “trends” that experienced traders are able to identify.

I’m looking to find a function that incorporates the sum total of a group of strategies, with a measurable variance. Since neural networks do this with a statistically significant increase in predictive accuracy, it tells us that there are such functions and patterns within the markets. Coupled with an expressive fundamental analysis, it should be possible to model the system with a lot more clarity than ANNs and handcrafted strategies do right now.

At the end of this, whether or not my hypothesis is right, I’ll have learned a lot.

As to backtesting, I know how to model my systems conservatively. Any backtesting system will have a parameterized setup, where I can change the spread, timing, and variance, to simulate things like “worst possible trading platform.”