Automated Trading Theory Discussion

permanentjaun · Sat, 10-21-2006, 07:01 PM

Hello all,

Here's the next of my open discussion threads. This time it is about automated trading systems. I'll start off by saying I do believe automated trading systems can work, but with help. In time I think I will be able to hand out business cards that title me as a Foreign Exchange Currency Automated Trading Systems Manager. Sounds nice right? So my belief is that automated trading systems can work, but will need to be monitored and tweaked to maintain efficiency and reliability.

This thread is to discuss why most automated systems fail however. It's to discuss their limitations and their abilities. Let me begin the discussion.

A computer can not make discretionary decisions. In my opinion this should rule out many indicators that need much discretion in making a decision based on them. Essentially, eliminate any indicators that do not follow price discreetly. RSI, CCI, MACD, stochastics, fibs, bollinger bands, nearly all indicators should be thrown out. Why do I say to eliminate just about every indicator out there? They don't follow price. They can form divergences. They can reach levels where their mathematical formulas are maxed out, such as in RSI, and can give neutral signals of being overbought and oversold. I believe it is asking too much of a computer to decide what to do in these situations.

Can a computer effectively recognize divergences? Chart patterns? It would add more to the confusion of the system when you must tell a computer by how much an indicator must have diverged before entering a trade. At what point does the divergence become effective in predicting a future market movement? If MACD diverges down for 2 periods is that a long enough divergence to base a decision? 5 periods? Perhaps only when it has diverged for 10 periods. There are many situations that must be accounted for. This is setting up for more possibilities for the system to fail.

Many times we look at several indicators and decide to throw out one of them since another indicator is giving more reliable signals. Can a computer decide which of these indicators to use? I will assume that to program a computer to do so would be too much of a project to profficiently carry out a trading system. There would be too many "what if" situations where you're only setting up the system to fail.

I will go so far as to say support/resistance levels should be thrown out of the equation. Get rid of fibs and other similar tools. There are plenty of times that prices explode through support/resistance levels easily. Even if resistance/support is confirmed as movement is stalled, the trend can continue through said levels. To program a computer to try and recognize support/resistance levels is just creating another "what if" situation for it to fail at. Can you correctly tell me why sometimes support/resistance levels are held while other times they mean nothing? If you can't do so with 100% accuracy, how can you tell that to a computer which has no thought process?

Thus, I believe that the way to program a trading system is to keep it simple. The system needs something that follows price solely since that is what we are concerned with predicting. We need something a computer can make decisions off of such that the indicator does not get maxed out, follows price, doesn't diverge, and is indiscretionary.

My conclusion is to use moving averages. They are based directly off of price, can not get maxed out, do not diverge, and computers can easily make decisions off of them.

Many will say that this is too simple to base a trading strategy off of. In fact, it can be as complicated as you want, for yourself. Do you create MA's off of the open, high, low, or close? Are they weighted, simple, exponential, triangular, etc.? Do you create it off of price closing above/below a MA or when another MA crosses? What period do you set the MA for? What time frame? I think those options right there are enough for you to think about. The point is that you create the system such that all the decisions were made by you. The computer can easily tell you when a cross or price movement will have triggered the trade. It doesn't need to recognize chart patterns, divergences, etc.. It simply needs to monitor price. The difficulty in deciding has been done by you.

permanentjaun · Sat, 10-21-2006, 07:02 PM

Now on to my thoughts on setting up the system.

We want a system that works well in any market, ranging and trending. To this I reply, a market is never ranging and a market is never trending. It is simply moving. I say this because what chart are you looking at to say a currency is ranging or trending? It has been shown that price movement slows down during times when major markets are closed and then explodes in movement when they open. Looking at a 10 minute chart would show many periods of inactivity or tight movement which could be categorized as ranging. Take a step back to a 2 hour chart or even a daily chart. Does the chart still show a ranging market? Conversely, look at a 1 or 2 hour chart. Does the currency look like it's trending? If so, take a step back to a weekly or monthly chart. GBP/USD has been trading in between 1.4000 and 2.000 since 1986. I argue that prices are just moving up or down. That is all we should be concerned with.

Many times I hear to exit a trade when the chart tells you to. This means don't set profit limits. If you exit with 10 pips be happy and don't force a trade to make 100. If a chart shows 100 pips then don't cut yourself short with 10 pips.

I'm an economist so a lot of how I think comes down to efficiency. To make it clear I'll describe extremes. You don't want to trade a daily chart with a 100 MA cross because it may be too slow. You don't want to trade a 1 minute chart with a 5 MA because it is too fast. Both situations present inefficient situations. In the first you're missing many opportunities to maximize profits and in the latter you're presented with too many signals which don't allow you to ride a price movement.

So my thoughts are that a chart needs to be used such that periods of inactivity are eliminated. If a price is "ranging" then its a large enough range that MA's can effectively keep up with price to play it just as equally as if it were "trending." This can be accomplished on many charts. We could use a 4 hour chart with a fast MA such as a weighted 5 MA. Would this give us too many whipsaws? We could also use a 30 minute with a 50 MA. Would this be too slow? That is where I believe the success of the system lies.

Find a MA that is overall quick enough to give a signal early, but not slow enough that the price movement is missed. It's not just the MA either. For the chart, is a 2 hour too slow such that each new candle allows too much movement so that when the next candle forms and you enter a position the run is already over? I'll present the converse of a 30 minute chart being too quick to enter a position.

permanentjaun · Sat, 10-21-2006, 07:02 PM

Lastly, I'll discuss limits. In my previous post I discussed how we should not cut our winners short. How would the system handle whipsaws or drastic movement against us to protect ourselves from profit loss?

One option I present is that the system is ALWAYS in the market. We want the chart that doesn't trade flat and is always going up or down. So if there is a cross to enter short we assume it will go down and when it crosses long we expect the market to continue going up. I assume this because why do we exit positions? We do so because we expect our profits to be maximized and further market movement is not strong. When we exit a long position its because we expect the market to reverse and head down. Would you not enter a short position if you trade a chart such that there is no sideways movement and only up or down? It doesn't make sense.

The system is based that we enter a long only in the belief that the market is going to go down.

So profits are protected by closing a position for only a few pips loss or even because of a whipsaw, and entering a new oppositie position. So if you entered long, protecting profits would involve a new position short. This allows us to maintain a position in the overall movement to make such losses insignificant.

What if the MA is too slow then? First off, we should be managing the system to eliminate this problem in the first place. Secondly, well I'm not so sure about secondly. I don't have a definitive answer to this problem. Certain ideas running through my head right now include exiting positions before the MA cross if price retraces 50% from the position entered in a certain amount of time. I'm taking suggestions on this part as well. Although, again, a lot of this problem can be minimized by finding the most efficient chart time and MA settings to make profits most efficient. It's sort of a 2 steps forward, one step back approach. We want to position ourselves to ride the 100 pip movement while taking small losses of 5-10 pips to guarantee ourselves a front row seat on the big ride.

Strategies could also include entering 2 lots. One that is exited with a profit of only a few pips to guarantee some pips. The other could be exited when the MA crosses or for an even close on the position. This would be difficult because of the typical problems associated with being "stopped out." Price can swing large enough to exit a trade, but not warrant the movement to reverse.

I'll conclude by summing up the problems of the system.

1. There is no clear profit protecting strategy. What happens when price reverses before the MA's can clearly tell us that? How do we protect ourselves?

2. Somewhat the same problem is telling the computer when to enter a trade. The computer would likely need a candle to complete and confirm the cross before entering the trade. Unfortunately what happens if we're looking at a 2 hour chart and the movement happens in that two hours then stops? We would have entered at the peak of the run and are vulnerable to a drop in price.

I believe the solutions are to finding the efficient chart intervals, and type of MA or MA's used. It needs US. We need to tell it what to do, not what it tells itself to do. Automated systems will need us.

Ok. Those are my thoughts. What do you all think? Thanks in advance. Matt

Sat, 10-21-2006, 07:01 PM	#1 (permalink)
permanentjaun Member Join Date: Dec 2005 Posts: 87	Automated Trading Theory Discussion Hello all, Here's the next of my open discussion threads. This time it is about automated trading systems. I'll start off by saying I do believe automated trading systems can work, but with help. In time I think I will be able to hand out business cards that title me as a Foreign Exchange Currency Automated Trading Systems Manager. Sounds nice right? So my belief is that automated trading systems can work, but will need to be monitored and tweaked to maintain efficiency and reliability. This thread is to discuss why most automated systems fail however. It's to discuss their limitations and their abilities. Let me begin the discussion. A computer can not make discretionary decisions. In my opinion this should rule out many indicators that need much discretion in making a decision based on them. Essentially, eliminate any indicators that do not follow price discreetly. RSI, CCI, MACD, stochastics, fibs, bollinger bands, nearly all indicators should be thrown out. Why do I say to eliminate just about every indicator out there? They don't follow price. They can form divergences. They can reach levels where their mathematical formulas are maxed out, such as in RSI, and can give neutral signals of being overbought and oversold. I believe it is asking too much of a computer to decide what to do in these situations. Can a computer effectively recognize divergences? Chart patterns? It would add more to the confusion of the system when you must tell a computer by how much an indicator must have diverged before entering a trade. At what point does the divergence become effective in predicting a future market movement? If MACD diverges down for 2 periods is that a long enough divergence to base a decision? 5 periods? Perhaps only when it has diverged for 10 periods. There are many situations that must be accounted for. This is setting up for more possibilities for the system to fail. Many times we look at several indicators and decide to throw out one of them since another indicator is giving more reliable signals. Can a computer decide which of these indicators to use? I will assume that to program a computer to do so would be too much of a project to profficiently carry out a trading system. There would be too many "what if" situations where you're only setting up the system to fail. I will go so far as to say support/resistance levels should be thrown out of the equation. Get rid of fibs and other similar tools. There are plenty of times that prices explode through support/resistance levels easily. Even if resistance/support is confirmed as movement is stalled, the trend can continue through said levels. To program a computer to try and recognize support/resistance levels is just creating another "what if" situation for it to fail at. Can you correctly tell me why sometimes support/resistance levels are held while other times they mean nothing? If you can't do so with 100% accuracy, how can you tell that to a computer which has no thought process? Thus, I believe that the way to program a trading system is to keep it simple. The system needs something that follows price solely since that is what we are concerned with predicting. We need something a computer can make decisions off of such that the indicator does not get maxed out, follows price, doesn't diverge, and is indiscretionary. My conclusion is to use moving averages. They are based directly off of price, can not get maxed out, do not diverge, and computers can easily make decisions off of them. Many will say that this is too simple to base a trading strategy off of. In fact, it can be as complicated as you want, for yourself. Do you create MA's off of the open, high, low, or close? Are they weighted, simple, exponential, triangular, etc.? Do you create it off of price closing above/below a MA or when another MA crosses? What period do you set the MA for? What time frame? I think those options right there are enough for you to think about. The point is that you create the system such that all the decisions were made by you. The computer can easily tell you when a cross or price movement will have triggered the trade. It doesn't need to recognize chart patterns, divergences, etc.. It simply needs to monitor price. The difficulty in deciding has been done by you.

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Sat, 10-21-2006, 07:02 PM	#2 (permalink)
permanentjaun Member Join Date: Dec 2005 Posts: 87	Now on to my thoughts on setting up the system. We want a system that works well in any market, ranging and trending. To this I reply, a market is never ranging and a market is never trending. It is simply moving. I say this because what chart are you looking at to say a currency is ranging or trending? It has been shown that price movement slows down during times when major markets are closed and then explodes in movement when they open. Looking at a 10 minute chart would show many periods of inactivity or tight movement which could be categorized as ranging. Take a step back to a 2 hour chart or even a daily chart. Does the chart still show a ranging market? Conversely, look at a 1 or 2 hour chart. Does the currency look like it's trending? If so, take a step back to a weekly or monthly chart. GBP/USD has been trading in between 1.4000 and 2.000 since 1986. I argue that prices are just moving up or down. That is all we should be concerned with. Many times I hear to exit a trade when the chart tells you to. This means don't set profit limits. If you exit with 10 pips be happy and don't force a trade to make 100. If a chart shows 100 pips then don't cut yourself short with 10 pips. I'm an economist so a lot of how I think comes down to efficiency. To make it clear I'll describe extremes. You don't want to trade a daily chart with a 100 MA cross because it may be too slow. You don't want to trade a 1 minute chart with a 5 MA because it is too fast. Both situations present inefficient situations. In the first you're missing many opportunities to maximize profits and in the latter you're presented with too many signals which don't allow you to ride a price movement. So my thoughts are that a chart needs to be used such that periods of inactivity are eliminated. If a price is "ranging" then its a large enough range that MA's can effectively keep up with price to play it just as equally as if it were "trending." This can be accomplished on many charts. We could use a 4 hour chart with a fast MA such as a weighted 5 MA. Would this give us too many whipsaws? We could also use a 30 minute with a 50 MA. Would this be too slow? That is where I believe the success of the system lies. Find a MA that is overall quick enough to give a signal early, but not slow enough that the price movement is missed. It's not just the MA either. For the chart, is a 2 hour too slow such that each new candle allows too much movement so that when the next candle forms and you enter a position the run is already over? I'll present the converse of a 30 minute chart being too quick to enter a position.

Sat, 10-21-2006, 07:02 PM	#3 (permalink)
permanentjaun Member Join Date: Dec 2005 Posts: 87	Lastly, I'll discuss limits. In my previous post I discussed how we should not cut our winners short. How would the system handle whipsaws or drastic movement against us to protect ourselves from profit loss? One option I present is that the system is ALWAYS in the market. We want the chart that doesn't trade flat and is always going up or down. So if there is a cross to enter short we assume it will go down and when it crosses long we expect the market to continue going up. I assume this because why do we exit positions? We do so because we expect our profits to be maximized and further market movement is not strong. When we exit a long position its because we expect the market to reverse and head down. Would you not enter a short position if you trade a chart such that there is no sideways movement and only up or down? It doesn't make sense. The system is based that we enter a long only in the belief that the market is going to go down. So profits are protected by closing a position for only a few pips loss or even because of a whipsaw, and entering a new oppositie position. So if you entered long, protecting profits would involve a new position short. This allows us to maintain a position in the overall movement to make such losses insignificant. What if the MA is too slow then? First off, we should be managing the system to eliminate this problem in the first place. Secondly, well I'm not so sure about secondly. I don't have a definitive answer to this problem. Certain ideas running through my head right now include exiting positions before the MA cross if price retraces 50% from the position entered in a certain amount of time. I'm taking suggestions on this part as well. Although, again, a lot of this problem can be minimized by finding the most efficient chart time and MA settings to make profits most efficient. It's sort of a 2 steps forward, one step back approach. We want to position ourselves to ride the 100 pip movement while taking small losses of 5-10 pips to guarantee ourselves a front row seat on the big ride. Strategies could also include entering 2 lots. One that is exited with a profit of only a few pips to guarantee some pips. The other could be exited when the MA crosses or for an even close on the position. This would be difficult because of the typical problems associated with being "stopped out." Price can swing large enough to exit a trade, but not warrant the movement to reverse. I'll conclude by summing up the problems of the system. 1. There is no clear profit protecting strategy. What happens when price reverses before the MA's can clearly tell us that? How do we protect ourselves? 2. Somewhat the same problem is telling the computer when to enter a trade. The computer would likely need a candle to complete and confirm the cross before entering the trade. Unfortunately what happens if we're looking at a 2 hour chart and the movement happens in that two hours then stops? We would have entered at the peak of the run and are vulnerable to a drop in price. I believe the solutions are to finding the efficient chart intervals, and type of MA or MA's used. It needs US. We need to tell it what to do, not what it tells itself to do. Automated systems will need us. Ok. Those are my thoughts. What do you all think? Thanks in advance. Matt