Nice algorithm. Seems to work pretty good. My suggestion is to use order_target_percent instead of value. This allows for you to earn exponentially on your positions resulting in more profit. Using the target percent order your algorithm achieved about 12% higher gain.

The algorithm performs very well except for 2011 and 2012. I'm not sure why those two years under performed.

Thanks Michael,

What's in my mind is that, this strategy will not work well or say maybe stagnant return during a sideway market (not quite sure that's the absolute good answer).

With that said, it is still possible to profit from sideway market as positions opened will be close immediately when price touches middle bollinger band, which is normally above the moving average 50 (entry price triggered for long positions).

Another thing in mind is that, I am still a beginner to algo trading.

Would anyone comment on the quality of the programming, could it be good enough to be used in a live trading environment? If not, why so.

Personally, I believe this code could be backtested with well return from the historical data perspective. But, putting it live trading would bring forth a lot of unforeseen issues.

In my mind, those issues could be:
liquidity / slicing effect / entry signal triggered could not be bought or fully filled
slippage
inconsistent price feed from broker ( possibly counterparty risk)
platform breakdown from the broker side

I cloned the algorithm and removed the commented out lines for easier reading. I also updated the moving average lines to use pandas rolling transformations instead of the built-in simple mavg transformation.

The built-in mavg function begins returning the true value only when the window is full. So for a 200-day moving average, this will begin returning the correct value on day 201. Here's a simple example to show the values for a 3-day moving average:

Day 1: report's the first day price
Day 2: report the average of the 2 day price
Day 3: report the average of the 3 day price. At the end of day 3, the moving average is now "filled" and will act as a true 3 day moving average
Day 4: Moving average of days 1-3
Day 5: Moving average of days 2-4

I can't comment on the recommendation to live trade this algo because we don't give out trading advice, but I would suggest to take a look at this thread: https://www.quantopian.com/posts/dark-side-of-technical-analysis

Looking at the backtest below, this strategy under-performed the market in 2007, hit a big position in 2008, and is riding those returns in subsequent years. If you are considering live trading an algorithm, I'd suggest to backtest it and try paper trading on Quantopian or using your IB paper account to see how it performs in the current market. Then if you're comfortable, you can choose to begin live trading.

Kian,
That's a tough question, is your code good enough to trade live? I'm also pretty new to algo trading, so I don't have any wisdom for you, just my perspective. As far as the strategy goes, your guess is as good as mine, so I don't have any advice there, but I can give you my objective opinion regarding code implementation.

I saw that you're only using the middle band of talib.BBANDS. The middle Bollinger band is a 20 day moving average (as coded), so your code can be simplified to use 3 different moving averages. Do you know that you're not actually using bollinger bands? I also found a couple redundant statements, which makes it seem like there could be holes in the analysis of the underlying strategy.

For algos that depend on moving average crossovers and other boolean triggers, I would suggest starting with the math and a visualization of the possible states your algo can take on. Representing the states as a matrix of boolean variables (or nodes on a graph) can help you simplify a strategy into a clear/concise form before coding it up.

My workflow is something along these lines:
1. Research
2. Backtest, if not satisfied, go to 1.
3. paper trade, if not satisfied, go to 1, or back to 2 and adjust.
4. Trade it live, if satisfied, it goes in the strategy portfolio, else, trash it and return to 1.
5. Re-assess
6. go to 1.

I'd be careful testing strategies that short stocks through the financial crisis because there was a period of time where short sales were either not allowed, or difficult to execute. If you are the only one allowed to short, you'll probably make a killing doing it. Your algo is a good example of that effect.

If you have not checked out zipline, Quantopian's backtesting engine, I recommend you do. Working with zipline will help you understand the tools you are working with. Because we rely on software to execute strategies, it's on you to play forward possible scenarios, and incorporate those views into your software. There will always be outside influences, like liquidity/slippage, but if you know how your code behaves (and that it works), you're probably heading in the right direction. It will also be easier to identify the issues you run into. Code does exactly what it's programmed to do. That means you need to thoroughly understand what you've written, if you are satisfied with the strategy, paper trade for a while, and then send it live.

Investing is risky, we can't eliminate the risk, but if it's managed intelligently, you can learn from mistakes and have the capital to trade another day. At the end of the day, we're all more or less feeling around in the dark, experienced traders might have a candle, but it's still a dimly lit room at best. Trading stocks successfully over the long haul requires sound strategy and a sense for the market, but it's always a learning process, the trick is to stay in the game.

I'd love for somebody with more experience to weigh in on this, otherwise it's just the blind leading the blind.

David

Kian,

Your strategy is well reasoned and certainly does well on that segment of SPY. So I tried it on a year-to-date segment of IWM during which that index performed badly. As you can see, I got radically different results. Possibly the fluctuations about the 0% level aren't handled well? Incidentally, SPY does very well on the same time segment. In any case, despite all the math and theorizing, this is a very, very empirical process that requires exhaustive testing against meaningful scenarios. Hope this helps?

Best regards,
Tom

Cant comment much on the coding, but it does seem that this strategy may catch some false breakouts , hence a stagnant performance in a non trending scenario. seems like if you caught a mjor uptrend, it would bank, but many strategies would do so, the key is in the ranging time frames, "how do you take advantage of those" type thoughts.