Is This Too Good to Be True??

[feng2088 10]

First of all I am a newbie on emini trading, but I have been learning, reading and programming my strategy. As I never scalped Emini on the10-tick chart before, I don't know if my strategy would work in reality, but it worked well on backtesting.

 

I use limit orders to buy and sell, and I also set a stoploss. So I can avoid most of the slippages. I understand that Stoploss is the only type of orders would have slippage among the three I use.

 

I backtested my strategy from 12/16/2010 -2/25/2011 on the 10-Tick chart. I got arond 80 to 200 trades a day. The profit factor was 3.7 and the percent profitable was above 80%.

 

I am doing fronttesting now and it's working well. I am going to add a max-loss-allow and if that number is hit, it will stop trading for the day. What else did I miss?

 

I am still afraid to go live yet. Some people said that you would get different numbers when you go live. Is that true? Thanks all.

[Tams 20]

First of all I am a newbie on emini trading, but I have been learning, reading and programming my strategy. As I never scalped Emini on the10-tick chart before, I don't know if my strategy would work in reality, but it worked well on backtesting.

 

I use limit orders to buy and sell, and I also set a stoploss. So I can avoid most of the slippages. I understand that Stoploss is the only type of orders would have slippage among the three I use.

 

I backtested my strategy from 12/16/2010 -2/25/2011 on the 10-Tick chart. I got arond 80 to 200 trades a day. The profit factor was 3.7 and the percent profitable was above 80%.

 

I am doing fronttesting now and it's working well. I am going to add a max-loss-allow and if that number is hit, it will stop trading for the day. What else did I miss?

 

I am still afraid to go live yet. Some people said that you would get different numbers when you go live. Is that true? Thanks all.

 

if it is too good to be true, it usually is.

 

what program are you using?

 

instead of jumping into autotrade,

I would suggest you to revise your code

to plot a green dot on the chart for every buy signal,

and a red dot for a sell signal.... and a black dot for exits.

 

you can then visually inspect the chart to see the plausibilities of such a strategy.

 

if everything looks ok, then go ahead to coding it in autotrade.

[feng2088 10]

Thanks Tams. I am using TradeStation and I have the dots plotted. They look good.

[Tams 20]

Thanks Tams. I am using TradeStation and I have the dots plotted. They look good.

 

excellent...

 

next step is to do forward testing.

 

Does TradeStation have papertrading accounts?

It would make testing a lot easier.

[feng2088 10]

i have been testing it with the old data and the current market ticks..I am able to set the initial capital , commission cost .. etc. Basically I am doing backtesting and forward testing together.

[zdo 20]

feng, good news and bad news

It is not too good to be true

AND

TS live results have a snow drop's chance in hell of being close to sim results.

 

Instead of drop dead going live, have you considered transitional going live? I've had good results in the past by first back testing to find if system has best times of day to trade and forward testing live for just those times and then gradually adding more live time, etc.

[feng2088 10]

Thanks zdo, is that always the case?? big difference between the live result and sim results....

[zdo 20]

Can't say it is always the case for everyone. But in every system I have ever tried that has similar properties to yours (fast timeframe, active, limit orders,etc), the results are different. Not necessarily big differences but always enough to make me look at my code some more... and the reasons are always strange and quircky

Hint: Some of the best work on these discrepancies is now 7+ years old in the TS forums

 

Have a great weekend all.

[UrmaBlume 10]

Thanks Tams. I am using TradeStation and I have the dots plotted. They look good.

 

The one big weakness with either back testing in TradeStation or using the TradeStation simulator is that with limit orders - if the price is hit you get a fill which is far from the reality in real-time, real-money trading.

 

If price just touches your limit order price once for a brief second your limit order will be filled and you will quickly learn that is hardly the case in real life. I know one beginner that was training on the simulator, making well over 100 round trips per session and always netting more than the daily range. Real time the same approach was a disaster.

 

The results from limit order back testing or limit order simulator trade in TradeStation are not only unreliable they can be very misleading.

 

None of this means your approach won't work but please be aware of the above and be careful.

 

cheers

 

UrmaBlume

[feng2088 10]

wow..Thanks UrmaBlume...I didn't know that. Is that because TS is too slow and they need to re-route my order from their server to exchange? will change my broker help?

[UrmaBlume 10]

wow..Thanks UrmaBlume...I didn't know that. Is that because TS is too slow and they need to re-route my order from their server to exchange? will change my broker help?

 

No, I believe that TradeStation is plenty fast for almost any human operated system and completely inadequate for anything approaching an automated high frequency system.

 

The condition only applies to back testing and the simulator. It has to do with the programming because in both the simulator and in back testing the software has no way of knowing your position in the queue for fills so it simply fills the order when the limit price is touched.

 

In TradeStation back testing and in their simulator your limit order is always first in the queue for fills. If that were true in real life anybody could kill the markets.

 

To verify what I am saying simply open the simulator place a limit order and you will see that the moment your price is touched you are filled. Place a real-money limit order and you will soon see that such "first touch" fills almost never happen and when it does is is usually because price ran through your limit price and in the wrong direction.

 

Another little know point about TradeStation back testing is that while minute and higher time frames can show fills inside the bar in back testing, tick, volume and range bars only store 4 prices, HLOC, so the software, during back testing can only see those four prices during testing.

 

The ten tick bar you reference is, on average, for less than 100 contracts which means there will be almost 1 bar per second during the average session, very fast for a human operated method.

 

Another caveat to your method is that you must achieve a very high hit rate when your profit target is less than half your risk/stop loss and even more so when you are paying retail juice.

 

Again none of this means your method is wrong but some years ago me and a couple of pals paid a hefty price to discover what I have mentioned here.

 

good luck

 

cheers

 

UrmaBlume

[feng2088 10]

This is extremely helpful...I will test it out with some real money next week and let you know how it goes. Thanks again

[feng2088 10]

Stupid question--- do you know if the order would be cancelled if it didn't get filled on that 10-tick bar? I don't want it to be filled 5 bars later

[UrmaBlume 10]

Stupid question--- do you know if the order would be cancelled if it didn't get filled on that 10-tick bar? I don't want it to be filled 5 bars later

 

In the simulator and in back testing it will fill on the first bar that touches the limit price, in real life there is no telling which bar it will fill on as that is dependent on both price and your position in the queue. In both the simulator and real life the order will remain in place until it is filled, canceled or the market closes.

 

I have tested such models that when automated and run on the simulator made well over 20-30 points per day, every day - but when run for real money - lost their ass.

 

The people that do this successfully don't use TradeStation, they are co-located and have the ability and capital to constantly "stuff" and cancel orders at size at several levels both above and below the last.

 

This kind of model is a very fast example of mean reversion trading and while a lot of people don't like the fact of it, it is what keeps our spreads so tight and benefits the retail trader.

 

 

cheers

 

UrmaBlume

[zdo 20]

UB, thanks for informing him of the issue with sim fills vs real limit order fills. I just assumed everyone knew about that aspect of the high hit rate systems...

 

feng, if you're staying with TS for this, one tech that has worked well for me with some systems is to run the system on a one tick chart (with setups signals coming in via ELC/ADE from your other 'time frame(s)') and go to mkt when price gets to or through where your limit would have been if you were still using limit orders.

Doesn't work for all systems but is generally a good way to test the viability of these 'too good to be true' systems. Slight drop off for some systems, which we just have to figure out a way to claw back some other way... hth

[UrmaBlume 10]

UB, thanks for informing him of the issue with sim fills vs real limit order fills. I just assumed everyone knew about that aspect of the high hit rate systems...

 

feng, if you're staying with TS for this, one tech that has worked well for me with some systems is to run the system on a one tick chart (with setups signals coming in via ELC/ADE from your other 'time frame(s)') and go to mkt when price gets to or through where your limit would have been if you were still using limit orders.

Doesn't work for all systems but is generally a good way to test the viability of these 'too good to be true' systems. Slight drop off for some systems, which we just have to figure out a way to claw back some other way... hth

 

The problem with the EL Collections/ADE toolset is that TradeStation still has a 1 minute time stamp so that collecting data, even in real-time from other time frame charts is useless and even sometimes "peeks ahead" for any time consideration less than 1 minute.

 

This problem is still relevant in TS 9.0 even with their new PSP (Price Series Providers) objects.

 

I don't know what is wrong with TS so that they bring strategy network when the tools they offer are deficient for strategy development and then bring these great data provider objects and smart order placement tickets and yet still operate in a 1 minute time frame in a millisecond world.

 

cheers

 

UrmaBlume

[feng2088 10]

Thanks both. To be safe, I might just change to a 1-min time frame. The theoretical return on other way is not as promising as I got from the 10-tick frame, but it's more realistic. The volume is pretty high on the 1 min chart. I am thinking instead of using a limit order to sell my position, can I set a profit target?? What is a chance of getting a slippage on that? Have you tried?

[UrmaBlume 10]

Thanks both. To be safe, I might just change to a 1-min time frame. The theoretical return on other way is not as promising as I got from the 10-tick frame, but it's more realistic. The volume is pretty high on the 1 min chart. I am thinking instead of using a limit order to sell my position, can I set a profit target?? What is a chance of getting a slippage on that? Have you tried?

 

The only way I know to semi-reliably test a system in TradeStation without using real money is to automate it with market entry and exit orders, write the stops so that they are conditional and when the condition is met a market order is placed and then run it in the simulator in real time.

 

The main weakness here is when a single transaction is much bigger than the bar size.

 

 

cheers

 

UrmaBlume

[Tradewinds 12]

When I first looked at a Zig-Zag study, I thought I was seeing a miracle. When looking at historical data, the Zig-Zag study was showing perfect peaks and bottoms. Then I found out that the Zig-Zag is delayed. It is possible to write a trading script that shows perfect entry and exit points if you are delaying the output. This is typically done with indexes that theoretically use price bars that are "in the future". But it is impossible to see into the future, so effectively what happens, is the plot is delayed.

 

It is possible to find leading indicators that tell you the price is about to turn. But it's always a matter of how probable it is.

 

Working with historical data is very different than trading live. The typical situation I like to point out is a bar or candle at a peak or bottom with a long tail on it. If you watch live trading, those long tails often happen so fast, that you have no opportunity to catch them unless you have an order in ahead of time. So the price did go to those price levels, but it happened in a matter of a second or two, and that is literally all the time you would have had to get a fill. So for practical purposes, it's very improbable that you could react fast enough. Again, if you had an order in ahead of time, and get lucky, then it works.

 

So, if your setups are not delayed, . . if they truly are leading indicators, and if there was enough time, and enough price movement to get a fill, those are the thing that make it actually trade-able.

[zdo 20]

Time stamp precision is not needed in all systems. There are also ways (via ELC or other objects) to increase the time stamp precision if necessary.

And even with the one minute restriction, on a one tick chart the data flow is serial ie the tick that just printed three ticks ago is not actually before the tick that printed four ticks ago - if that really matters.

UB, With the systems you run, I understand your need for a higher level of precision (and have wondered why you're still with TS... )

Long ago I commited myself to accept less accuracy and honesty in the data stream - a tick and any aggregate of ticks is still only a symbolic representation of the actual 'auctions' occurring...

I work to prepare to thrive in even more 'chaos' than we have now and in several orders lower of data completeness... I think I could be up and running in a day or so if I were dropped into any exchange traded market on the planet... large or small

 

...hey guys, in the TS forums this archeological dig has already been taken down to the level where you can just brush off the top dust to find the goodies ... still time consuming but better than waiting here in TL for it to show up piecemeal...hth

[jtrader88 10]

TS backtesting only has price data and doesn't consider how many bids/asks there were at that point in time. I tried autotrading in TS and made money 90% of the time in Sim Mode. Lost money in real trading. I think a slower system looking for big moves where you wouldn't mind a couple of ticks of slippage might work though. 80 trades/day won't work in real trading.

[zdo 20]

I think a slower system looking for big moves where you wouldn't mind a couple of ticks of slippage might work though. 80 trades/day won't work in real trading.

 

jtrader88 Welcome... hate to give you such a hard ass reply to your first post on TL ...

but

Wouldn't it be more accurate to say something like 'Most 80 trades a day systems won't work in real trading' ?

Maybe you gave up too quickly. If that is the op's nature then a "slower system looking for big moves" would ultimately Not work at all for him - even if it's profitable.

If nothing else will do, the op can find a way...

[UrmaBlume 10]

TS backtesting only has price data and doesn't consider how many bids/asks there were at that point in time. I tried autotrading in TS and made money 90% of the time in Sim Mode. Lost money in real trading. I think a slower system looking for big moves where you wouldn't mind a couple of ticks of slippage might work though. 80 trades/day won't work in real trading.

 

While most systems of any kind don't work in real-money, real-time trading, certainly there are systems that make hundreds or even thousands of transactions per session that do work and make a lot of money for the operators. It is called HFT.

 

These systems not only work they do well over half of the trade in almost every liquid stock/future.

[TimRacette 11]

One issue I could see is that by back testing or using a paper account you are often able to buy at the bid and sell at the ask. This is not always the case on a live account because in order to get a fill you'd need the mark to be on your bid or below to get a fill (slightly confusing), that being said I have a strategy that has an 85.7% first target hit rate. i.e. I sell half my position at this first target and it gets there 85.7% of the time. I also have a colleague who runs a hedge fund and had only 5 down days in 2010 (no they didn't wipe out his whole year, he was still +5M on years end) so it is possible.

[Tradewinds 12]

I think that extremely successful strategies are possible. I also think that there are incredibly few people who come up with them. Out of all the competitors, somebody will win the gold medal.