Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
@SMJCB Sorry, I haven't referred to what I do. I do not fully follow your line of questions. I was just discussing what I thought should be an obvious or self-evident fact. There are many more complex ways of making speculations. For example, one can use comparative reasoning or logic. As an example, we have a few cases where the price of Bitcoin has rallied when a country's currency has had trouble and/or government. One can use such comparative logic to look for countries who might experience currency crisis and buy Bitcoin/short the currency. Of course, this specific example has been a local phenomena and probably not trade-able. We've also seen that developed markets tend to be mean reverting. We can thus speculate or watch for new emerging markets to become more mean reverting. Comparative logic can also be used to reason about, price action or other types of things.
What I think that some of the best speculators do is they identify the factors (real factors) that trader's use to make decisions and form hypothesis about what is likely to happen and then they also understand price patterns (empirical patterns) and combine those insights to produce superior trading. I agree with Kevin that if you go to far to the right-hand side "too speculative" that it becomes more difficult to be consistent. In other cases, great speculators seem to be good at understanding when the market has mispriced an event. In any rate, I agree also it is extremely difficult to profit even from great predictions..
I'd rather shift this focus back to systems which is what Kevin does. I only interjected to provide my opinion, really fact, on what I thought should be obvious.
Anyway, there is also potential value in making your trading system output a prediction, or qualify score, versus a binary signal. Dr. Howard Bandy gave me some insights into this, and I have taken a liking to the concept. Let's just imagine you build your trading system to produce a signal. As part of your optimization process, you will make "local optimal' decisions as you optimize/craft your systems. At any rate, if you take the statistical/empirical view then each trade will have the same expectancy regardless of whether you take one trade or all the trades given the assumption that your system's trades are independent and don't rely on outlier profits. So, let's imagine you have a small account (like most new/small traders). What's likely to happen? You might build a few profitable systems and then lose interest as you won't have capital to deploy to even the profitable systems you have. Now, if you make your systems output predictions versus signals you can sort/rank the various signals and or do other sorts of interesting things. Of course, you need to find/define the quality factor. But, there are many sorts of things one can do with a score or quality factor (or prediction) that cannot be applied to a binary signal alone.
Anyway, Kevin have you ever used this idea of outputting a prediction or confidence value for a system or set of subsytems? I suspect you prefer to keep things simple and trade each system and track it. But, do you see anything wrong with the concept?
Can you help answer these questions from other members on NexusFi?
One thing I always tell people is that regardless of your method, style, testing approach (or lack thereof), the market is the ultimate judge.
In the end, whatever process one uses to trade should be validated through real money results. That feedback loop is critical. Listen to the market, and if what one is doing is not working (is not profitable), then that is the cue that you should change something.
Kevin, when you give a trade idea a quick first test - as I understood your webinars, you backtest a simpistic "system" with this entry and check the equity curve. What exit(s) do you use for that? Just a timed one or something else?
Maybe you answered it elsewhere - I haven't found it yet.
Thanks for the question. I usually use a time based exit (exit after X bars) and maybe a stop loss.
Many times, if this looks good, I actually just keep the simple exits, and not try to improve things with more complicated exits.
I should point out that testing an entry with simple exits makes the assumption that the entry is what really is important, and that is not necessarily the case. Many times the exit is more important, or even the interaction of the entry and exit is what really matters.
These days, I probably do more of entry and exit testing at the same time, because of the interaction effects...
As I thought that I learned a lot, I wrote down what I learned and was suprised how short the boil-down about the exits actually was. I'd divide them roughly into
hard (initial stops and predefined targets)
timed (after x bars; at noon)
anxious/greedy (first probitable open; when x% of profit are eaten back)
trailing (from parabolic to trend reversal)
Am I missing something big in this area? Would you add any class or a representative exemplar to my list?
Hi Kevin: I have a general question about backtesting on different bar types. Let's say I run a system in real time on a renko bar type for a week (I understand that a week isn't statistically significant - it's just to get a general idea). At the end of the week I compare the results to Strategy Analyzer simulator results. I then calculate the slippage between the simulated and live results. In order to use the slippage factor, I add/deduct the slippage average to the profit target/stop loss. Do you see any problem with that going forward? For instance, on the UB if I have a slippage average of 2 ticks and my profit target is 6 ticks, I would reduce my profit target to 4 ticks for live trading. Stop loss is always 4 ticks. I felt this was a more accurate way to calculate slippage rather than using the slippage from the Historical Fill Processing option. I've traded live using this method and it seems to be okay as long as I get filled at the entry price (this is a problem sometimes in bonds).
I appreciate your input on this. Thanks.
Also, I execute manually. This is not intended for automation. The goal is to estimate the most likely successful target for this system.
Thanks for the prompt reply. I trade Jigsaw DOM and order flow but I'm experimenting with strategy development in Ninja.
When I said "Live", what I meant was letting the strategy run in real time to see what the difference was between live data and backtest data. I also test it in replay. The replay data matches the real time data very well. I understand that this doesn't take into account fills, etc. The strategy executes at market and exits on a limit order.
My goal for doing this exercise was to normalize the backtest data so that I know where I have to make adjustments for real time trading. The backtest engine in Ninja has a lot of inherent problems making the results unreliable. IMO it's really useless. But my thought was that if I could normalize the data the same way you do in simple normalization statistics, I could factor in the adjusted values when I run a test in replay. Then do manual walk forward testing every week to see how the real time data holds up to the replay data where the strategy would include the factor as slippage.
I've attached a couple of jpgs as examples. The first one has two charts: The one on top is backtest from a chart - not Strategy Analyzer - for today's ZB session. The chart below is the replay data for the same. The second jpg is replay data for the strategy with a 2 tick slippage included which turns out to match the replay data, at least for today.
After testing different sets of profit targets/stop losses, the results suggested that 4/4 were the best inputs to use - i.e. most probable successful profit target with a matching stop loss.
I just don't trust the backtest engine. That's why I came up with this scheme to see if I could come up with a valid, reliable factor to use with an inaccurate backtesting engine. I still have to calculate the number of outliers over the range of available data.
I'm really just trying to work with a defective backtest engine to get reasonably reliable results.
I know you use TradeStation. I can program in EL. I switched to Ninja because I liked the graphics. I'm not that great at programming in Ninja though. Trading order flow I really didn't need TradeStation any more.
Anyway, just wanted your opinion on this. I don't want to waste a lot of time with Ninja programming if I can't acquire reliable results.