Speaking of jokes I am not so sure this is caused by data, it could be that one test used bars with zero volume in them and the other did not. Always something!!!
It would be interesting to know if the above testing results were carried out concurrently, with all set-ups/ variables/ factors equal, specifically for your particular evaluation purpose.
1. Total system trading ideology without any specific disclosing details Total idea that every reasonable hypothesis will provide better operation only in specific market conditions is not new. For example, a big money move can take 1 â 30 minutes. So, if we could detect some way this movement in the first 5 minutes, we have some probability to pinch off a little bit from this big money. Another situation when all market parameters say that situation is bad and the market, apparently, are going down and no reason for fast changing market outlook. Choppy market has the specific character and signature revealed in everyone accessible digital form; letâs play this state by shorting local heights. I believe, you could add your favourite strategy in this list. But most of you are trying to use a position enter signal only inside trading vehicle: e.g. if you trade NQ, you analyze tick-by-tick, bid and ask size changes or just 1 minute bar. Someone uses just temporal filters like do not trade on Thursday or situational filters like FOMC announcement times. Nevertheless, if temporal or situational filter feeds from real time any kind of market data flow, it will be very close to that I am talking about. Only remember, I consider the trigger signal produced from outside NQ data, if I trade NQ (see Figure 1). Strategy set can include n-listed trading implemented ideas; each of them is triggered by according market state acceptable or suitable for specific strategy realization. When âStrategy 1â gets order for execution, âPosition entering algorithmâ, as shown in Figure 1, will seek time and place to take the prescribed position. If this algorithm cannot take good enough position, it can cancel the execution order. If the position was successfully taken, then immediately stop and target orders are placed. âStop order correction and support algorithmâ together with âTarget order correction and supportâ accompany the position till one of them has to be executed. âPosition closing algorithmâ completes this trading line. Report generation means provides me all necessary hardcopy and log-file materials. I am not with my computer during trading hours. Computer sends me the emails to my cell phone when any position was taken and final trading result. System provides an alert signal if something happens. I communicate with my trading computer through the remote desktop only if I have the alert signal. -------------------------------
One way over overcoming false signals through bad ticks is by using some weighted calculation of price and vol across a specific time period. This won't eliminate them completely but will half the number of false signals. I process price and volume off each tick but I look for a specific confirmed movement in the price volume information across a 5 or ten minute period, so one bad tick will change the calculation slighly but not nessesary trigger a false entry. The funny thing is I didn't design it this way, I just noticed after examining false ticks on the chart - working with tick data has its advantages too. Runningbear
Runningbear, thank you for the response. I have tested the tick-filtering procedure like you described. The funny thing is I didnât design it this way too. The situation with the false ticks is not so simple. In the reality different data feeds give different real time values for bid, ask values and sizes, and last price action and volume. I will talk about it later. The described here system does not fight possible false ticks, system accepts every tick as real one. I include all possible false ticks into a tick noise function. Better way when used system reveals robustness to known type of the tick noise. If you know this noise distribution, it is possible to calculate a possible system behavior with the predetermined confidence interval. Row count gives that system may be successful if the noise STD smaller than at least 0.2-0.4 target average range. That means, if system trades ES and constant noise is around 1-2 ticks, the system can be used for targets not smaller than 5 ticks. -------------------------------
Stephencrowley, sorry this thread will not include millisecond timescale trading results. I think the system development for millisecond range can take much more time and money. ----------------------------------
"I include all possible false ticks into a tick noise function." ------------------------------- What is tick noise function' I couldn't find such function in TS2000i. how about coding that if price spikes up or down Unusal X% in either direction, wait for X seconds to see it price hold itself near that range, and then only initiate a new trade
The noise function is the distribution of the tick divergence in size or value between two data feed sources. I found that more useful for the model development and testing a transfer noise function which is the distribution of differences in bar values (open, high, low, close prices, e.g. for 1 min bar) between real time received values by used trading platform and feeding broker database. For example, if I use IB and their API with my custom developed platform, like it or not, I am loosing 10-20% ESâs ticks. Nana trader, I am not sure you could find such function in TS2000i. Thank you for the advice. This idea will be hardly useful in 15-200 minutes trading timescale. -----------------------------------------