How to reduce slippage at entry and exit

[mastermind007]

Hi

Do a back test to check on your system/afl to see how many false signals you get if your signals triggers few seconds before the close of the candle.

i.e check for the signals reversing in last 5/10 seconds.

Now compare the cost of this to actual slippage / impact cost for your trades.

Choose the better option and be happy with it :thumb:

Happy

Happyji

This suggestion will only work in theory and not in practice.

To arrive at realistc approximation of slippage impact, you need to run this over large volume of data over few years.

Also, the data for that period will need to be in ticks of seconds which will become too big to manage.

On a 32 bit machine, you can store max to max 2 months and on 64 bit machine, limit is somewhere near 6 months.

Also, you will need few different AFLs that can handle this multiple scenarios

 

[Amit Nanda]

Hmmmm

So, effectively, You are relying on a crossover. There are two things you can do. Your primary source of problem is delay in time and slippage in price is symptom of that problem.

One thing that you can do is instead of relying on Close, rely on High for entry.

This will eliminate repaint of signal and will allow you to enter within a second of price moving above the horz line but this approach will increase chances of getting trapped in false breakouts and your calculation of levels will have to somehow factor that away.

Yes, I had back tested this option as well, that is, enter/exit a trade as soon as the price touches the threshold (rather waiting for close of the candle). There are periods when the number of false signals increase manifold; thereby rendering the strategy useless. On the other hand, there are many periods where it works really. Overall, on back testing for the last 5-6 years, this option turns out to be highly inconsistent in the long run. It would have been a good solution had number of false signals been less and/or the profit curve over 5-6 year period were smooth/consistent.

I am not sure what else can be done in this case

 

[Amit Nanda]

Happyji

This suggestion will only work in theory and not in practice.

To arrive at realistc approximation of slippage impact, you need to run this over large volume of data over few years.

Also, the data for that period will need to be in ticks of seconds which will become too big to manage.

On a 32 bit machine, you can store max to max 2 months and on 64 bit machine, limit is somewhere near 6 months.

Also, you will need few different AFLs that can handle this multiple scenarios

That's true. Moreover, I do not have second level data.

 

[Happy_Singh]

That's true. Moreover, I do not have second level data.

Was suggesting compare with your actual slippage and not trying to derive with tick data etc...

The coding will be only limited to look into the trades triggered in last 5 seconds before the close of the candle
and assess how many of them (%) reversed back / vanished at the close of the candle.

Essentially you re code the system to trigger 5 seconds before close and compare it with current results.

Step 2 is compare this impact cost with your Actual Slippage you are getting now.


Happy

 

[Amit Nanda]

Was suggesting compare with your actual slippage and not trying to derive with tick data etc...

The coding will be only limited to look into the trades triggered in last 5 seconds before the close of the candle
and assess how many of them (%) reversed back / vanished at the close of the candle.

Essentially you re code the system to trigger 5 seconds before close and compare it with current results.

Step 2 is compare this impact cost with your Actual Slippage you are getting now.


Happy

Yes, I get that. However the historical data that I have is in 1 minute format. To test what you have suggested, one needs data at 1 second interval, which I do not have. Is there a way I can get historical data at 1 second interval for past 5-6 years?

 

[Happy_Singh]

Yes, I get that. However the historical data that I have is in 1 minute format. To test what you have suggested, one needs data at 1 second interval, which I do not have. Is there a way I can get historical data at 1 second interval for past 5-6 years?

Ok, that's a valid concern

1) Try out the experiment with a gap of one minute (this won't directly extrapolate to lesser duration of 5 sec, but it can give you an outer limit)

2) Look for sample data of few months and test it (this can extrapolate to bigger set of data, and give you a clue about order/percentage)

3) Forward test it along with your trading (You will definitely arrive at a conclusion within few months, btw you have to do this any which way before you go live)

Happy

 

[Amit Nanda]

Ok, that's a valid concern

1) Try out the experiment with a gap of one minute (this won't directly extrapolate to lesser duration of 5 sec, but it can give you an outer limit)

2) Look for sample data of few months and test it (this can extrapolate to bigger set of data, and give you a clue about order/percentage)

3) Forward test it along with your trading (You will definitely arrive at a conclusion within few months, btw you have to do this any which way before you go live)

Happy

Thanks. I will try this. Changing the AFL for point 1) above would be tricky. Would give it a shot.

 

[mastermind007]

Yes, I had back tested this option as well, that is, enter/exit a trade as soon as the price touches the threshold (rather waiting for close of the candle). There are periods when the number of false signals increase manifold; thereby rendering the strategy useless. On the other hand, there are many periods where it works really. Overall, on back testing for the last 5-6 years, this option turns out to be highly inconsistent in the long run. It would have been a good solution had number of false signals been less and/or the profit curve over 5-6 year period were smooth/consistent.

I am not sure what else can be done in this case

I already told you the answer. The logic that determines the entry level has to be optimally adjusted to minimize the false trades.

Let me try it in different words and instead of slippage, i will use word skid. Its a rainy season after all and 2 wheelers are skidding all over the place.

If you use close-confirmation logic, you will have vertical skids (on price axis).
If you use High / Low to fire entries, you will have horizontal skids (on time axis)

Earlier, if you were adding say 1 point to your compute level to weed out false breaks, you will need to add 2 or maybe 3.

Other thing you can try is that High crossing above needs to have achieved at least 80% of average ADM

 

[Lalpha]

Try putting in predictive entry values with a stop limit type of an order, buy above / sell below. For ex if you want to take a long trade, at 09:55:55 use the available data until that point of time and compute what is the minimum value required in the next minute at which your trigger can get generated and place that order at 09:55:59.