What is Orbit Products?

What are Orbit Products?

MoA’s Orbit Products are the first flagship decentralized financial products MoA introduces to the world. They are managed by Orbit, our machine learning based investment model built to automatically execute investment decisions. Orbit offers the following three major features that distinguish it from other financial products.

Orbit has three distinguishing features.

Orbit has three distinguishing features

Machine-learning based that detects strong trends

Our model is machine learning based and it automatically considers a vast number of potential scenarios for a given market environment. The only human intervention occurs in the early stages when information is fed to the model. It tests all the potential investment scenarios for each asset to find the optimal one. It uses machine learning validation techniques to avoid over-fitting. Once the testing and configuration is completed, Orbit autonomously determines whether to buy, sell or hold an asset.

24 hour, 365 day risk management

Orbit secures the portfolio for 24/7 to protect profits against sudden and significant market moves. Around-the-clock risk management is a fundamental part of Orbit as the market for digital assets is extremely volatile: one sudden drop can wipe away a significant portion of profits, and even threaten the capital. Orbit offers an automatic system giving portfolios a trailing stop protection. As Orbit monitors the market 24/7, it simultaneously monitors the market for promising investment opportunities.

Transparent performance display via blockchain

Orbit’s performance will be recorded onto the Edenchain blockchain to maintain transparency. The core of all financial activities is taking actions today for what one believes to happen tomorrow. As our model is in the business of making the best educated guess for what cannot be guaranteed as of the present, it is crucial for us, the developers behind the model, to transparently display the records we used to have the model arrive at its conclusion. By doing so, we minimize uncertainty for the investors and earn trust from them. Blockchain is a perfect fit for this.

1. Machine-learning based that detects strong trends

Our model is machine learning based and it automatically considers a vast number of potential scenarios for a given market environment. The only human intervention occurs in the early stages when information is fed to the model. It tests all the potential investment scenarios for each asset to find the optimal one. It uses machine learning validation techniques to avoid over-fitting. Once the testing and configuration is completed, Orbit autonomously determines whether to buy, sell or hold an asset.

2. 24 hour, 365 day risk management

Orbit secures the portfolio for 24/7 to protect profits against sudden and significant market moves. Around-the-clock risk management is a fundamental part of Orbit as the market for digital assets is extremely volatile: one sudden drop can wipe away a significant portion of profits, and even threaten the capital. Orbit offers an automatic system giving portfolios a trailing stop protection. As Orbit monitors the market 24/7, it simultaneously monitors the market for promising investment opportunities.

3. Transparent performance display via blockchain

Orbit’s performance will be recorded onto the EdenChain blockchain to maintain transparency. The core of all financial activities is taking actions today for what one believes to happen tomorrow. As our model is in the business of making the best educated guess for what cannot be guaranteed as of the present, it is crucial for us, the developers behind the model, to transparently display the records we used to have the model arrive at its conclusion. By doing so, we minimize uncertainty for the investors and earn trust from them. Blockchain is a perfect fit for this.

Investment Philosophy

Investment Philosophy

Orbit’s investment philosophy is based on trend-following, a well-known trading strategy beloved by investors for many decades. Trend following is known to play out in various markets not limited to crypto, including stocks, commodities and so on, as it works stably across asset classes and time frames. Owing to its flexible application, Two Centuries of Trend Following remarks that trend following has been a persistent feature of all the financial markets for centuries.

Aggregated performance of the trend
 Aggregate performance of the trend on all sectors, sharpe ratio = 0.72, Two centuries of Trend Following

The core idea of trend following is this: buy an asset when it shows a positive trend, then sell it when the up-trend is over. David Richardo, a legendary political economist, nicely summarizes the concept as follows: “cut your loss and let your profit run”. It is an intuitive and simple concept that has worked for centuries.

A simple schematic for the Trend Following concept
A simple schematic for the Trend Following concept

One of the biggest assumptions in the strategy is that a trend, once it has formed, has a proclivity to prolong the dynamic over time. In other words, a real trend does not end over a short period of time, and the phenomenon is easily observed when one looks at historical financial price charts. It works well for trend followers as it removes the burden of having to accurately predict the future price of the assets under consideration. One needs to simply jump on an up-trend and ride it until the trend ends.

Orbit's Core Strategy

Orbit's Core Strategy

Orbit’s investment strategy can be boiled down to three points: finding a trend, riding it, then exiting when the trend is over. To best execute the strategy, we need to answer the following three questions: 1) which asset to buy, 2) when to buy, 3) when to sell.

 

Orbit’s main strength lies in answering the second question, “when to buy”. It distinguishes itself among competitors by detecting early signals for strong trends to maximize the profit. 


Early trend detection is crucial in trend following for it is closely related to your ROI and risk management. This is especially true with highly volatile assets, namely crypto assets, where seeing a 30% price drop in 15 minutes is very possible. For a market as highly volatile as digital assets, investors have difficulty using traditional trend detection methods. Orbit answers the call for a more sophisticated approach targeted for crypto markets, and indeed any assets with extreme volatility.

Orbit and Science

Orbit and Science

To a certain degree, the current data for the value of an asset reflects its future value. Why? Because active decisions made by investors in the present time are affecting the future value of the asset. The idea of the Efficient Market Hypothesis (EMH) supports this, implying that the price of an asset is supposed to move around the intrinsic value of the asset with added noise. For instance, if the true underlying price of an asset (the price that reflects its intrinsic value) is 100, the price can be any of 95,105,103, 98, and so. Unless there is a special momentum, the price will revolve, or orbit around, a consensus price, i.e. its intrinsic value. This is similar to the natural behavior of electrons found in an atom.

diagram showing quantum theory

In quantum atomic theory, electrons exist in specific regions around an atom’s nucleus called orbits. Different orbits represent different energy levels, and an electron resides in one orbit at a given time depending on its energy level. What is notable is that an electron jumps, and does not slide up, to the next level only when sufficient energy is applied to the electron. The behavior where it requires a “certain” level, also referred as a “quantized” amount to cause an action, is the core of quantum theory.

This atomic model is a good analogy that explains how our Orbit model works. Orbit’s underlying assumption is that “an anomalous event will likely precede a new price discovery”. Like electrons in an atom, price movements should show a strong enough anomaly to escape its previous pattern; otherwise it will fail to escape from its current behavior. An anomaly, by definition, possesses pattern-changing dynamics, therefore, it is supposed to be an extreme event that shows a unique pattern very rarely observed. If we can detect such anomalous events in price movements, we can identify the formation of a strong trend in its early stage with high probability. Therefore, MoA’s Orbit has a sophisticated mechanism for anomaly detection for strong and early trend forming.

Orbit’s Anomaly Detection
Mechanism Catches the Early Trend

Orbit’s Anomaly Detection Mechanism Catches the Early Trend

The biggest technological strength of Orbit is clearly its ability to detect anomalies found in price movement. Orbit thoroughly studies the price data for possible anomalous events that are likely to form strong trends. If Orbit detects an anomaly that presents with high possibility for an upward trend, it concludes that one should buy the asset and ride the trend until it is over. This is the heart of Orbit.

anomaly in graph
Anomaly detection technology identifies events with a rare occurrence rate in the given data set.

It sounds easy enough at first. However, defining “unlikely events” in numbers and clear criteria to teach software is not an easy task at all.

To solve that problem, Orbit uses the following definition: if expected data and actual data show a certain degree of differentiation, the actual event is an anomalous event. Every asset has its own price movement distribution, and we can model the distribution using machine learning to estimate its price movement. Afterward, we can compare the two sets of data to examine the degree of differentiation of the two. If a big enough differentiation is detected, then we can regard the data point as an anomaly. 

Correctly teaching the model the underlying data distribution is enormously important as it will project the market movement based upon the lessons it has received. To teach the model balanced and accurate lessons, Orbit uses MDN (Mixture Density Network) for data distribution learning because MDN can handle multiple data distributions. As a result, Orbit is able to make market estimations based on precise data distribution learnings.



What is Mixture Density Distribution
Mixture Density Network (MDN) for anomaly detection

Once the model learns a price movement data distribution through machine learning, then it can identify anomalous events by comparing the locations of actual events against the estimated distribution.

How to tell anomaly in a graph
Anomaly event is detected by comparing estimated data against actual data
Anomaly is found outside of the range
Events that fall within the noise range are not anomalous

Orbit has a machine learning based mechanism for anomaly detection; when it receives price movement data, it estimates the probability for the presence of anomalous events. Orbit makes a decision based on that information.

Risk Management

Risk Management

The biggest pitfall in trend following is getting false signals, as only around 20% of signals form contiguous and profitable trends. This explains why risk management is a key component in the application of the trend following strategy. Without a good risk management plan, funds will be depleted from following false signals before it gets to enjoy profits.

- Why are there so many false signals?

There are multiple reasons as to why forming a stable and long lasting trend is difficult. For one, there is strong tension between the Bulls (or optimists who believe that the intrinsic value of a given asset has changed) fighting against Bears (or pessimists). The trend will not form until Bulls take over the majority of the market atmosphere. If the asset has a large market capitalization, forming a strong positive trend requires more investors, making the upward trend that much more difficult to form. In the beginning only a few optimists will think the current price is cheaper than its intrinsic value and will buy it. To form a trend, more buyers need to expect upward price actions, otherwise the trend will not set. Most trends fail to overcome the threshold. As it is inevitable for many trends to die out in the early stage, it is also unavoidable that many of your attempts to ride the early trends to fail .Therefore, you need to have a strong risk management plan to minimize losses while waiting for a sustaining trend to ride out .

- Timing the selling is a complex art

More complicated than buying is selling (exiting a trend) as it is a more complex and subtle process. Selling too early results in losing out on the opportunity for higher profit, while selling too late may jeopardize unrealized profit. Selling is truly an art of striking the right balance between one’s greed and security.

- Trailing stop mechanism that works

Orbit has a trailing stop mechanism for risk management. A trailing stop places a stop order (automatic liquidation) activated when the preset parameter is met. If the stop price is reached, the sell order will be automatically executed. A trailing stop is a proven tool to protect profit.

What is a trailing stop

Orbit uses anomaly detection mainly to check entering time, and uses trailing stop to time the exit point. When new price data is fed into Orbit, two algorithms, one for entering, the other for exiting, calculate its condition, to make a final decision.

Backtest Results

Backtest Results

We tested Orbit products multiple times with historical data under a plethora of conditions to thoroughly assess characteristics, including performance, trading frequency and maximum loss of those products. When developing a new model, backtesting is the only available option for assessment, and it has its pros and cons. One must carefully interpret the test results as there is no guarantee that a model that worked in the past will also work in the future. Testing only with historical data can lead to overfitting of the model, and it is one of the most common and serious problems in building models, one that is also not easy to overcome. 

To avoid overfitting while developing Orbit products, we used two strategies: 

  1. Intentionally created mutations in Orbit and conduct tests, 
  2. varying the conditions used for data testing.

As a result, backtesting results we presented here are not based on a single model but encompass a set of distribution metrics collected from varying mutations of the parent Orbit model.



Backtest results
Table for Orbit’s Backtesting Results

The table indicates that precision is stable because standard deviation of precisions is 0.094. If the model’s precision is above 0.247 the system will generate profits because the mean for the return is positive(0.001516).

The following scatter chart shows a characteristic of the orbit’s variants on its precision and machine learning score. Precisions means precision for entering, exit points. Score represents the degree of precision of the machine learning algorithm. Higher precision and score indicate high trend detection success. If the Orbit model is being overfitted, the scatter chart is likely to show no pattern because it is generated from randomness.



Scatter chart displaying Orbit’s score on the precision and machine learning result
Scatter chart displaying Orbit’s score on the precision and machine learning result

Below chart shows correlation between score and expected return on single trading. If the average return is above 0, it is likely to generate profits as the number of trades is increased.

Scatter chart displaying correlation between score and expected return on single trading.
Scatter chart displaying correlation between score and expected return on single trading.

The below chart shows entry and exit points generated by the selected orbit model. Green means buy, black means sell. The green triangle is generated by anomaly detection whereas black is by trailing stop.

Bitcoin market tested with Orbit model
Chart from Orbit’s backtesting result showing entry and exit points for Bitcoin market

The below chart shows the portfolio value of the above model. Interesting point on the chart is that there is no severe drawdown period even in the 2018 crypto winter season. The model seems to properly manage the risk.

Graph shows no MDD
Orbit increased the portfolio value of the above case with no MDD.

Get early access to the latest updates on the MoA Investment Solutions