Well, after much testing and verification. The solution is that it works, multiple uncorrelated strategies on the same instrument do, in fact, smooth the equity curve in my testing. Of course, I'm sure this all depends on what strategies you choose to use and how you use it. Like they say, a tool in the wrong hands becomes dangerous. It also helps to "interconnect" the strategies to specifically avoid correlation. This is just to thank those who were honest and accurate about this point. Hopefully, to save anyone the trouble of asking, it's impossible to divulge any further details about this for intellectual property reasons. Sincerely, Wayne
By treed, I think he means think of your strategies as a "tree". Make certain strategies decisions depend on others in such a way to make certain they never correlate. Wayne
You are just guessing and your guess is oversimplified and largely biased by your work on "interconnected" strategies. The only one who can answer this question is acrary himself.
For all interested in system development I've opened a blog at: acrary.wordpress.com Nothing there yet, just starting to figure out how to do it. I'm doing a blog so I can edit posts and correct mistakes. I also don't spend much time on the net nowadays and I wanted to write my posts offline and post them when they look good. Anyway, I'll discuss various topics of research and trading there. Not promising anything and will only post when I have some time to do it right. Hopefully, I'm not violating the tos here by posting this. Good trading to all.
Alan, Thanks for your ideas you post. They sparked renewed energy. I built my own framework in C# from scratch to implement the ideas. One of the biggest ideas was to focus on regular periodic returns rather than trade returns. I also use monthly. To date, I now have a system (can this be right??) that makes profit 100% of every month in the last 5 years. It reads "infinity" for several of the stats. The Sortino and Sharpe are off the charts. The annualized return is over 400%. It turns out the trade win/rate is over 99% too. I don't know if you mentioned it but one concept I added into the mix of using portfolio trading is fuzzy logic. In other words, the primary models heavily use variable positions sizes instead of black/white long and short positions. Right now, I'm wrapping up the finishing touches on the API to connect to the broker. I'll trade it on a demo account for a few weeks before going live. Anyway, I'd love to read your blog and post for your feedback (as you have time) because there's so many ways, I'm sure, to tweak these models and even add more models. It seems like a whole world has opened up. Wayne
Before anyone makes the comments about commission and slippage. It accounts for commission and, while it trades with the prevailing trend, it always trades counter to the current momentum using limit orders so it doesn't suffer slippage. It makes 20+ pips average on over 5,500 trades in 5 years using tick data. All the stats on this strategy look great bar none. The only potential downside (which I'm working on ironing out). Since the markets don't follow a Gaussian random distribution but instead have "fat tails", it means that a couple dozen trades in the 5 year period go haywire and drawdown 100 times more than the average trade before finally making a profit. In the big picture, the Net Profit / Max DD looks awesome. So even with this "glitch" it can make phenomenal profit just by using those outliers as the risk in money management. Still it appears they occur at more or less regular intervals corresponding to economic reports. That's the area of my statistical research at present. So some more analysis may allow the model to avoid trading at those higher risk times (or alter the algorithm at those points to profit with less DD). One most fascinating thing about the markets that can make you a fortune once you get your mind around it is this. They are 98% predictable! It's true. The predictability lies in it's randomness. It sounds contradictory. But consider this. You can count on the market behaving truly randomly 98% of the time. That's fairly amazing. This model for example experiences more risk in the 2% or less times when the market gets off the randomness and skews seriously bullish or bearish without any pullbacks, rallies or retracements for much longer than ordinary. All you have to do is build a model that capitalizes on the random distribution of prices to have trades win 98% of the time. Unfortunately, I don't think 100% accuracy is entirely possible due to the fat tails in the random distribution. Think about it, could you make a strategy that will make money even if you feed it random generated data? If not, you'll always be fighting an uphill battle. The set of models described above profits closer to 100% of the time on random data generated to simulate the market. But only 99% or 98% on real historical tick data. When you think about it, that's pretty amazing that the market is so predictably random, isn't it? Why doesn't it skew to a fixed pattern more often with so many people wanting to follow trends or channels? It now almost looks to me also like some major players intentionally force the randomness to make it harder for the crowds. It's just too incredibly consistent. If my account size and volume were large enough I think it would tend to enforce the randomness since it uses the randomness to profit. It's impossible for me to believe that I'm the first or only one to figure this out. It's just too terribly predictable. Sincerely, Wayne
I've noticed very similar things. I've also found that these fat-tail periods are best identified using proxies to psychological indicators... at least if you are trading a more macro product (sector etfs, et cetera)... Even measuring a recent 'risk/reward' metric versus historic 'risk/reward' gives a fairly good indication whether the market is currently experiencing herd behavior...