any thoughts on these indicators?
- Thread starter user268
- Start date
To the point, stop wasting time.
From what I've learned here and in my own attempts with indicators, here are a couple of things I've learned for evaluating indicators, I'll try to summarize some of the main points:
- does it make sense? there are lots of ways to derive an indicator, doing all kinds of clever arithmetic and including many different points of data. But ultimately does the end number from all of those inputs and calculations represent something that passes a common sense approach of it's meaningful, or some coincidence from a bunch of noise. A simple example -- a long bar with high volume; plausible as it could represent a lot of people going in that direction. A simple example -- a long bar on a day where the digits of the day if summed individually equal 23; there was a movie about this, the main character saw sums of 23 everywhere, but it really exceeds common sense believability. (of course, these common sense views may or may not be true). Example, I had one that calculated off of 4 bars, and predicted a large move; I later estimated a move of that size would usually take a period of time comprising 100 bars or more; this seemed unbelievable, no way 4 bars could predict the next 100. If it were more like 4 bras predicting the next 14 bars it would have been more believable, maybe even up to 30. But to 100, it seemed more like this would have been random chance. (of course, my beliefs here may or may not be real or true.)
- ask the question, then try to disprove with events in the data. The opposite is looking for events in the data, then trying to figure out what the indicator did just before them, and then drawing conclusions. It's really easy to look back at the peaks and troughs, then match the indicator's behavior just before that. But this won't work in real time going forward. It's better to ask, 'does the blue line crossing 30 upwards precede a peak?' then test against the data to get an objective estimate. Most backtesting software will do this, my favorite is Forex Strategy Builder. You will get an estimate of '30% (or 50% or some%) it precedes a peak', then you have to do more work to see if this is random chance or significant.
- better to use objective numbers for levels or signals than subjective descriptions of 'it's turning down' or 'it's really high'. In real time going forward this won't work very well. When you watch the indicator at the right, it's always turning up or down or really high or low. It's better to ask, 'when it's over 40', or 'when the red line is 10 above the blue line', or similar. Next, if it really does precede a top or bottom, next have to figure out the expected time range of the top or bottom; if you don't know if it's in 1 day or 10 days, it might not work out so well.
- the previous 2 points lead to choosing parameters, and cautions about curve-fitting or over-optimzation. Any indicator can be made to fit any past data with the right parameters, but then it won't work well going forward.
- be sure the indicator uses data available at the time of calculation. Sometimes it may include a data point that could not be known at the time of calculation, and it won't work well in real time going forward. An example is the 'repaint' issue.
- how does the interpolation affect it? if calculating based on Closes, for example, that is one point out of the whole day, throws out a lot of data. If the interval shifted (ex, use 12:00 as close instead of 4:00), how would this change the signals? I prefer highs and lows, like the Donchian Channels, because they are a bit more robust against this issue. Some people may consider this insignificant and nothing to think about.
I find the most use of an indicator is transforming the noisy price data into a series of single numbers the computer can read. (Some people will argue here that O-H-L-C-Vol are the purest and best single number series.) Then, how to figure out if it is giving signals above what would be expected by random chance. Then, if it's consistent across different samples of data.
But, don't take my word for it, I don't know much, and I cannot argue to defend the above points. They are bits and pieces I've accumulated here on ET, hopefully they provide a key word to search for more information, or point out something to look for. There are lots of threads from posters smarter than me who are really good at developing systems. If we're lucky, maybe some folks will note some of their favorites. One of mine was 'do you see patterns in random walks?'. I think another one very popular is "system development with acrary".
Anyways, I hope my post is helpful, not just listing basics that everyone knows already, good luck with your indicators, I hope they work out well for you, if nothing else I always find it fun to write the code for them.
- does it make sense? there are lots of ways to derive an indicator, doing all kinds of clever arithmetic and including many different points of data. But ultimately does the end number from all of those inputs and calculations represent something that passes a common sense approach of it's meaningful, or some coincidence from a bunch of noise. A simple example -- a long bar with high volume; plausible as it could represent a lot of people going in that direction. A simple example -- a long bar on a day where the digits of the day if summed individually equal 23; there was a movie about this, the main character saw sums of 23 everywhere, but it really exceeds common sense believability. (of course, these common sense views may or may not be true). Example, I had one that calculated off of 4 bars, and predicted a large move; I later estimated a move of that size would usually take a period of time comprising 100 bars or more; this seemed unbelievable, no way 4 bars could predict the next 100. If it were more like 4 bras predicting the next 14 bars it would have been more believable, maybe even up to 30. But to 100, it seemed more like this would have been random chance. (of course, my beliefs here may or may not be real or true.)
- ask the question, then try to disprove with events in the data. The opposite is looking for events in the data, then trying to figure out what the indicator did just before them, and then drawing conclusions. It's really easy to look back at the peaks and troughs, then match the indicator's behavior just before that. But this won't work in real time going forward. It's better to ask, 'does the blue line crossing 30 upwards precede a peak?' then test against the data to get an objective estimate. Most backtesting software will do this, my favorite is Forex Strategy Builder. You will get an estimate of '30% (or 50% or some%) it precedes a peak', then you have to do more work to see if this is random chance or significant.
- better to use objective numbers for levels or signals than subjective descriptions of 'it's turning down' or 'it's really high'. In real time going forward this won't work very well. When you watch the indicator at the right, it's always turning up or down or really high or low. It's better to ask, 'when it's over 40', or 'when the red line is 10 above the blue line', or similar. Next, if it really does precede a top or bottom, next have to figure out the expected time range of the top or bottom; if you don't know if it's in 1 day or 10 days, it might not work out so well.
- the previous 2 points lead to choosing parameters, and cautions about curve-fitting or over-optimzation. Any indicator can be made to fit any past data with the right parameters, but then it won't work well going forward.
- be sure the indicator uses data available at the time of calculation. Sometimes it may include a data point that could not be known at the time of calculation, and it won't work well in real time going forward. An example is the 'repaint' issue.
- how does the interpolation affect it? if calculating based on Closes, for example, that is one point out of the whole day, throws out a lot of data. If the interval shifted (ex, use 12:00 as close instead of 4:00), how would this change the signals? I prefer highs and lows, like the Donchian Channels, because they are a bit more robust against this issue. Some people may consider this insignificant and nothing to think about.
I find the most use of an indicator is transforming the noisy price data into a series of single numbers the computer can read. (Some people will argue here that O-H-L-C-Vol are the purest and best single number series.) Then, how to figure out if it is giving signals above what would be expected by random chance. Then, if it's consistent across different samples of data.
But, don't take my word for it, I don't know much, and I cannot argue to defend the above points. They are bits and pieces I've accumulated here on ET, hopefully they provide a key word to search for more information, or point out something to look for. There are lots of threads from posters smarter than me who are really good at developing systems. If we're lucky, maybe some folks will note some of their favorites. One of mine was 'do you see patterns in random walks?'. I think another one very popular is "system development with acrary".
Anyways, I hope my post is helpful, not just listing basics that everyone knows already, good luck with your indicators, I hope they work out well for you, if nothing else I always find it fun to write the code for them.
No, it is actually the response you wanted to "hear".