WaveTek is an API using Pattern recognition on 1-minute price patterns. We use web-sockets to stream data to you, after our Deep Learning Model processes the Top of the Book orders.
As data arrives from IEX it gets processed by our Deep Learning model. We stream out pattern recognition results which include next minute OHLC and Volume estimated values. This includes directional changes to the price as well as estimated pattern duration.
If you’re an algo-trader, developer or otherwise engage with expert advisors, indicators or trading bots, WaveTek provides some very valuable insights:
Is a momentum pattern unfolding? if not, then maybe you shouldn’t open a position.
Do you have an open position but don’t know when to close it? WaveTek will give you frequently updated estimations on reversals or end of momentum, so you can make better informed decisions.
Are you building an Indicator or EA? Don’t rely on lagging TA indicators, use the power of AI pattern recognition to improve your indicator’s or EA’s performance.
The frequency and speed at which a lot of those patterns unfold is phenomenal! Whilst longer patterns may take 5, 6 or even 10 minutes to unfold, very often some violent and large swings occur in just 2 to 3 minutes.
How confident are you that you can watch, open a position, watch closely its momentum, and then close it when the time is right?
It is no surprise that the vast majority of retail traders, more than 90%, lose their money when intra-day trading. And that’s usually with trading on 5-minute, 15-minute or 30-minute intervals. The fact is that such low latency and high frequency trading is done by bots and algorithms.
Very accurate. Surprisingly accurate, considering that the vast majority of algo-traders and bot coders are skeptical of Machine Learning and AI, because of how often they fail to implement AI and ML in a useful manner. WaveTek took close to 3 years to get to this level of accuracy and a serious amount of grey matter.
But without further ado, here are the actual results as it currently stands:
Actual Pattern as it unfolds, produces an R2 Score of 0.94. Meaning that WaveTek has a very high regressive ability to detect new emerging patterns.
Estimated prices (OHLC + Volume) for the following minutes is actually a predictive regression, meaning that WaveTek can estimate how the next minute OHLC + Volume will move. Whereas predicting the future is impossible, predicting how a momentum pattern will unfold isn’t. The R2 Score here is 0.85 meaning that WaveTek can quite accurately tell you at what price level is the pattern likely to move at, including both lower and higher levels which imply directional changes.
Directional changes, e.g., reversal or continuation as a binary classification task is slightly harder. F1 Score (combined precision and recall) on future price direction currently sits at 0.80.
Estimated duration of the pattern, e.g., is the momentum likely to keep unfolding or will it end soon is reflected in the R2 Score. A return array of zeroes (OHLC + Vol) means that the pattern will estimated to end at that point.
The WaveTek technology doesn’t come without caveats, and not mentioning them here would be insincere. After all, our goal is to help the algo-trader and developer improve his or her profit margins, so you need to be aware of the caveats.
First and foremost, previous training data does not guarantee future performance.
Meaning that WaveTek has been training on patterns from 2019 until 2022, including the bubble and the collapse of it. However new patterns might emerge, which WaveTek is not trained on. We will mitigate this risk by frequently training WaveTek with new data so that no such performance drift takes place.
Shorter patterns are harder to predict than longer ones.
This might sound confusing, but here’s the gist of it; shorter patterns (e.g., 2 minute or 3 minute ones) are more violent, more abrupt, more noisy and therefore more unpredictable. However, their directional prediction is actually more accurate, as they tend to follow the direction towards which they manifested.
The opposite also holds true; longer patterns are more accurate to predict (higher R2 scores) but their directional change accuracy becomes harder to estimate, meaning that as the momentum exhausts itself, directional changes are harder to predict.
We cannot predict patterns before they have manifested at least partially
Unfortunately for us, for a pattern to be appropriately recognized, part of it has to already have unfolded. Meaning that we cannot predict the future, and therefore we cannot predict when a pattern will emerge.
There’s a variety of reasons why patterns emerge;
A bot placed a large order, and other bots responded to that
A human trader placed a large order and other traders responded to that
Some news, gossip or information is leaked and traders perceived that event as a reason to buy or sell
any other reason you can think of.
What this means is that, at least part of the pattern must have manifested itself, at which point WaveTek can indeed detect it and begin making predictions about it. Which brings us to the last caveat:
You can, but probably shouldn’t use WaveTek to open positions
Opening a position implies you’ve detected an event, or a reason to do so. Using WaveTek, a pattern must have partially emerged before it begins to make accurate predictions.
So while in theory you can use WaveTek to infer if a position should be opened, in practice you may find that by the time you’ve opened a position you should close it. The timeframe to do so may range from a few minutes to a few milliseconds. So be prepared that if you use WaveTek for opening positions you may be closing them too fast, and therefore not making a profit. If your broker charges you for doing so, then you will incur charges which could have been avoided.
In short, any of the following:
Close positions
Keep positions open
Detect trading events, which manifest as price patterns
Scalping and ultra-low-latency trading
Develop indicators and Expert Advisors
Data-mine analytics and insights
Yes, all of the below:
WaveTek is more accurate and robust than any EA or TA indicator, because it doesn’t rely on lagging information, but refreshes with current new data
However, WaveTek is not a magic bullet, there’s still room for error and you should plan for it
WaveTek performance and accuracy will keep improving as we add more training data, it is currently trained on about 7Gb of data, but Deep Learning models are well known to perform better as more data and more samples are added
WaveTek is near real time, meaning the delay is in the millisecond region, as such you can use it in a much faster pace than any of the TA indicators or EAs.
WaveTek is currently trained only on IEX equities. If you would like WaveTek to work with Cryptos, let us know.
Feel free to contact us for anything that relates to WaveTek and its perfromance. Enjoy and have fun.