I see two charts on your past performance one have not make any money in last few months, other shows volatility of flip the coin buy/ sell trading model. Are you for real ?

MARTIN SMIETANSKI - I'm glad you had a chance to review our web site. The two graphs on our site are updated weekly. The first one was used for 'proof of concept'. After turning $10k into $112k in around 6 months (with virtually no drawdown), we knew our algo was working. Around October of last year, we decided to take on an even bigger challenge, which was to create an algo-driven subscription service that any trader could follow intuitively and do just as well. The second graph shows the last 60 days and, yes, it might resemble a coin toss, however, that is largely due to unprecedented, erratic fundamentals (i.e. the Trump effect). Behind the scenes, we are testing and tweaking live on multiple fronts as we hone in on a set of parameters and a framework for our soon-to-be launched subscription service. The level of analysis is pretty complex and involves a lot of computational math, probability modeling, design, and some pretty heavy duty coding. More at RagingFX.com

Hi. Sorry to bring some realism here, but your curve is terrible, there is an early 20% drawdown after just a couple of months of being live! and there is no way that first curve traded 9 times a day, there are week long flat periods in there. If this is a signal service then you have to be aware that no manual trader will keep putting on trades through a 20% drawdown. That curve is not suitable for the type of business you want to run.

You may want some unbiased testing done on your algo; my former employer offered that service though it's a bit pricey and you won't be allowed to advertise with the testing results. But if you're serious about learning whether your algorithm is any good, you should consider such a service.

Have to agree. This does not look decent. The last 60 das are sideways - I can live with that. But the longer test chart (which still is pathetic in terms of timeframe) really shows a strategy that is not working anymore - the later half of the chart is basically flat. Even the detailed report has no sensible length of testing - what about you run those strats in a backtest for 5 years and check the result. Looks too much like random noise. I particularly have how the 1000$ provit between may and Sept 2016 turned into basically a flat line.

Neil Crossman - No doubt the drawdown during our proof-of-concept phase appears steep. We were trading an average of 25 to 30 signals daily back then, which would explain the 9.4 signal average you questioned. It has since settled down considerably to between 0 and 10 daily signals, a range more suitable for a subscription service. You're welcome to review our process and data during this discovery phase at RagingFX.com/report.html - look for the following link: 'Click here for a Detailed PDF Report'.

The PDF Report shows the results from 4 servers set up to benchmark their respective performance data. After about 4 months of trading with our algo, we chose the server with the highest accumulated profits and began tweaking it for a subscription service. At this juncture, our goal was to achieve similar win/loss ratios but with fewer daily signals, a tall order if you've ever tried a similar task. After nearly 5 months of tweaking, we are encouraged and ready to launch on March 27.

Roger Darin - Thanks for the advice.

We considered 3rd party testing but felt that our business model would suffice. With RagingFX, every day, every subscriber can see what was traded and how it compares to the signals they received from us the day before. Since the shelf-life (after entering one of our signals) is limited to a maximum of 31 hours, socializing our 'testing' on a daily basis is very doable and hopefully a big selling point.

With third party testers, the results are only as good as the time when they were tested. One would have to assume that future trades would render similar results, which we all know is a normal disclaimer in this business for statistical reasons. On the other hand, having every subscriber check for performance errors on a daily basis, we believe, could close the gap of 'uncertainty' while potentially upping the ante.

Of course, after we launch, we will know better.

I'd be most interested in your opinion or anyone else. Are we dead wrong?

Are you using any machine learning algorithms based on non-parametric Bayesian statistics?

Personally, I think that in terms of return the performance is quite good. It's about 115K $ profit for an investment of 10K $ (money in the account) + 23K $ (the "reserve" for the potential drawdown) = 33K $. That is something in the range of 350% return in less than one year, if I have correctly understood. The problem is that it may have been a lucky strike, and thus I agree with Thomas Tomiczek, the overall timeframe is too short to give any real confidence, especially considering the shape of the equity line. Maybe to publish some backtesting results with details on the criteria used for these tests may help to better evaluate the quality of the algo.

why wouldnt you use your own capital or start a hedge fund if returns are so good with amazing drawdown you claim ?

Carmen-Gabriela Stefanita - in a nutshell... We looked at applying non-parametric Bayesian linear regressions but decided on taking an approach less dependent on streaming data and more on overlaid proven strategies with streaming data clusters. When we find a match, we apply our probability factors to an optimization model, which does a comparative analysis on 8 different data sets from 8 different proven strategies. We then select/construct a signal that meets about 30 points of preset, strategy-dependent, indicator criteria. We developed various custom machine learning components, which we apply both in batch and in real-time, These ML components are distilled to one numeric value, which are used to benchmark a market sentiment index. In the event that our market sentiment index exceeds either an upper or lower preset limit, this factor will override our calculated ML benchmark and allow approved signals with lower probabilities of success to pass through.

Paolo Villani - I appreciate your feedback. I can see that our graphs are creating some confusion, since they combine our 'proof-of-concept' phase along with our subscription service design. It would be like demonstrating a racing car fit for an Indi-500 race to an audience looking for wheels to get around. I can see the disconnect and will address it prior to our launch on March 27.

Another point of interest... From my 20 years experience with analyzing data, I've never been a proponent of 'back testing'. Everything we have done here has been 'forward' testing. To me, it would be like learning how to drive from a video and believing that you are now a competent driver. I realize the industry expects it but in this case, as with much of everything we have done so far, we have chosen to buck the trend. Forward testing tends to push us to our limits within a realm of relevance. Knowing that our algo is performing with today's data, just makes selling the service that much easier. JMHO

Piyush Sheth - Great question! This project began as a wager with two top traders in London and morphed into what it is today. It has and continues to be an algo for my own personal trading. However, the only way to go from a good solution to an amazing one is to involve subscribers. These are individuals that we don't know personally but who, by their mere involvement, force our team to push the limits on our combined talents. I used this strategy with another software product I developed, and it worked really well. It was the most efficient and direct way, I know, to commercialize a software solution.

Since our personal trading needs are perfectly aligned with the successful performance of our service, we believe the service will sell itself via word-of-mouth (social media, of course) based on its consistent performance. There's no rush in getting it wrong. Eventually we hope to interest institutions who may ask us to private label our service for their specific needs.

Great explanation. Thank you!

Nice one Tom.. good luck with this project looks good!

Tom, Upon looking at the return stream, I see that the stream has a VERY distinct pattern, inherent in the May-Feb full set and the Jan-Feb subset holding periods. Think fractal and self similarity. The dampening of profits since Aug is due to a major regime change, which the system is unable to adjust.

First, congratulations on finding a regime, which you have learned to successfully capitalize. However, your next task is to identify the phase changes leading into and out of THIS regime, which is VERY clear in the return stream. The return stream will give you clues of HOW to identify when the regime is changing. Also, this task will help decrease the drawdown magnitude and duration.

Tom Kadala - Per your response to Piyush Sheth question, be aware that once you go down the commercial route of a subscription based model, it will be challenging - on several fronts - to market the model to institutions. Happy to share thoughts offline. Good luck .

James Stedman, Ph.D - Many thanks for your thoughts. The greatest challenge we have had so far is replicating our own trading success to the masses via a subscription service. It feels similar to training a robot as one's own human replacement. ...very challenging but fun too! The details and nuances are many, and not necessarily all algo-related. We have had to overcome so many hurdles to get as far as we have that we are truly encouraged that we can make this work, but like everything else in life that is worth considering, it takes a little time and patience, a little experience, and a little luck along the way.

First things first. We our launching our service March 27. With an uptime near 99% for the past six months or so, we feel we are ready to move beyond beta and up the ante. Six months from now and assuming we achieve 90% replication (even 80%), we'll be ready to speak with institutions. Your input would be most appreciated, now and then. Please reach out to me at your conve

Tom Kadala Will do so. Success is a journey, enjoy the ride!