The trading of traditional financial and cryptocurrency markets is executed at a disadvantage for the vast majority of participants: retail traders. Access to traditional financial markets for retail traders is dominated by brokerages that aggressively profit from their clients through artificially adjusted market prices, spreads, unrealistic fees, and commissions, in many cases unbeknownst to the trader. Further to this, crypto buyers and sellers pay maker, taker, and leverage fees to trade cryptocurrencies and are unable to leverage their crypto to trade traditional financial markets.
Traditional brokerages and centralized cryptocurrency exchanges that offer margin trading prey on retail traders worldwide by forcing them into a disadvantageous position against the market. Retail traders buy and sell cryptocurrencies, fiat pairs, stocks, and futures at less favorable prices, paying various commissions and fees, while having zero market edge when deciding what to buy and sell. Traders are obliged to access the markets via brokerages and exchanges whose interests are misaligned with their own, if not in direct opposition. To make matters worse, there is still more value being extracted from retail traders' transactions. Their trading data finds its way, on a regular basis, into the hands of hedge funds and banks who use it to build profitable trading strategies that are in direct conflict and therefore detrimental to retail traders.
Quantfury ends these practices once and for all. It makes the trading of traditional and cryptocurrency markets free, fair, and transparent. Users trade at the real-time prices of the major cryptocurrency and traditional exchanges without any commissions, leverage fees, maker and taker fees.
Table of Contents
Trade All Markets with Unmatched Conditions
List of Asset Types Traded
Trading Platform Architecture
Tokenizing Quantfury’s Economy
Monetizing Trading Data
Quantfury Advantage Matrix
Notes and Further Reading
Quantfury offers traders, investors, and cryptocurrency holders, to trade traditional financial and cryptocurrency markets with unmatched conditions.
For Cryptocurrency Holders and Retail Traders
• Trade free of commissions, leverage fees, and maker/taker fees;
• Leverage cryptocurrency to buy and sell equities, cryptocurrency, fiat pairs, and futures;
• Always trade at the best (lowest) available ask price and the best (highest) available bid price.
Crypto exchanges only offer cryptocurrency margin trading while quoting synthetic prices, and charging unnecessary fees, putting their user base at a consistent disadvantage.
Traditional retail brokerages bet against their users and profit from the artificially adjusted market bid and ask prices, trade commissions, and leverage fees.
All retail traders execute their buy orders at the current exchange sell price and their sell orders at the current exchange buy price; for this reason, they are automatically in a losing position at the time they entered the transaction. Simply put, traders always buy higher and sell lower than the last market price of the asset they trade. When various trading platforms and brokerages adjust the spread between the buy and sell prices to make them even wider, charging higher trade commissions and higher fees for leverage and withdrawal, then the edge against the trader will grow exponentially, and trader losses will become greater. These practices represent the core of the business model adopted by the majority of traditional retail brokerages and crypto exchanges offering margin trading. This business model also is known as the “churn and burn” model, where user retention becomes a lesser priority because it is understood that users will eventually lose a substantial amount of their funds (‘burn’) and stop trading (‘churn’). The goal is to burn as much of the trader's capital as possible before the churn occurs.
Although the “churn and burn” model can explain the depth of the user’s losses, it still cannot explain why such a high percentage of retail traders lose money. The fact that the majority of traders lose money is public knowledge (1), yet the industry guards the exact statistics of those losses. Some of those statistics are available from time to time due to some of the retail brokerages having to publish their trader performance reports as a regulatory requirement. These reports provide few details, purposefully so, yet still point to the fact that 65% to 80% of traders lose money (1). Research work conducted by academics from around the world analyzing retail trading for stocks, futures, and currency pairs points to a much bleaker picture of trader losses. That research shows that 95% of traders lose money, while 4% break even and 1% percent turn a profit. In addition, it takes on average three years for a losing trader to quit speculating the market and stop trading (2).
A trader’s decision to buy or sell specific financial instruments is not only based on readily available market data, such as the asset price, but also driven by behavioral patterns. These patterns specifically relate to the decision making of the majority of retail traders based on the potential value of losses and gains, as well as evaluating their losses and gains using some heuristics (3). In cognitive psychology and decision theory, loss aversion refers to people's tendency to prefer avoiding losses rather than acquiring equivalent gains: for them, it is better not to lose $5 than to gain $5. The principle is very prominent in the domain of economics and trading. What distinguishes loss aversion from risk aversion is that the utility of a monetary payoff depends on what was previously experienced, or was expected to happen. Some studies have suggested that losses are twice as powerful, psychologically, as gains. In fact, a recently published MIT research article (4) analyzing trade data from more than 80,000 traditional retail brokerage trades concluded that indeed the majority of losing traders held on to their losses twice as long as winning traders.
Quantfury will not change the percentage of retail traders losing money due to still the existing difference between the real-time market buy and sell price, while market psychology will continue to be present, however, the size of users’ losses will be significantly curbed.
Trade All Markets with Unmatched Conditions
The Quantfury app is the most cost-effective, fair, and transparent choice for trading the markets.
Quantfury has developed a smartphone mobile app, which allows cryptocurrency owners to trade through fiat currencies financial instruments, such as, cryptocurrencies, fiat pairs, equities, and futures listed on U.S. and European exchanges, without paying a commission, leverage, or other types of fees. All financial instruments that are traded are always quoted at real-time exchange buy and sell prices (5), which are never artificially adjusted.
Only cryptocurrencies (no “fiat” currencies) are accepted to fund the user account balance wallet to begin trading on the Quantfury app. The users access constant trading power in USD 20x their selected value of account balance funded in cryptocurrency equivalent. This trading power remains constant. The account balance secures potential losses incurred in the fiat currency of the trading power according to the current exchange rate as the user trades. Users are not permitted to incur losses greater than the account balance value according to the most recent exchange rate data. The maximum size of trading power allowed by the Quantfury trading app is $200,000 which requires $10,000 of account balance funding (in crypto equivalent).
A user funds their account balance wallet, in the form of Bitcoin, equivalent to $100, and receives a constant trading power of up to $2,000 to use as they choose to trade both cryptocurrency and traditional financial markets. This trading power of up to $2,000 is available to allocate into various trade positions and in various amounts, always stays constant regardless of losses incurred from trades. The losses incurred from trading cannot exceed the fiat equivalent of the user's account balance value; otherwise, trading will be stopped and positions liquidated.
List of Asset Types Traded
Equities and ETFs
475 US and EU listed mid and large-cap stocks and ETFs
BTC/USDT, ETH/USDT, EOS/USDT, LTC/USDT, DASH/USDT, IOTA/USDT, NEO/USDT, XMR/USDT, XRP/USDT, ETC/USDT, ZEC/USDT, ADA/USDT, QTUM/USDT, XTZ/USDT, LINK/USDT, DOT/USDT
S&P 500 E-mini, Dow E-mini, NASDAQ-100 E-mini, Crude Oil, Natural Gas, Gold, Platinum, Silver, Copper
USD/CAD EUR/USD EUR/GBP USD/JPY GBP/USD AUD/USD NZD/USD USD/CHF EUR/JPY CAD/JPY GBP/JPY AUD/JPY NZD/JPY CHF/JPY
Trading Platform Architecture
The Quantfury platform uses a microservices architecture to deliver next-generation scalability for its trading app user base. Its platform sets the standard for reliably scalable and high-performance data delivery for transaction-driven services, breaking the work into multiple loosely coupled services that can be independently upgraded and scaled in multiple dimensions.
The Quantfury trading platform is designed to match user trades internally and provide continuous market making activity that delivers guaranteed execution of any user trade order that was not matched. The Quantfury platform is the industry's only “light pool” available for retail traders direct access. The Quantfury platform is the total opposite of a “dark pool” concept which exists primarily for institutional traders. In Quantfury's light pool every participant gets unmatched industry trading conditions, all participants are retail traders, and all trading platform activity can be audited and verified through the blockchain. Anybody can be the owner of the pool, and benefit from its operations through token ownership. Therefore the interests of participants, token holders, and the Quantfury platform itself are always fully aligned.
Quantfury guarantees all its user trades are quoted and executed at the exchanges’ best bid or ask prices, without any trade commissions or leverage fees, regardless of the number of daily active users.
The Quantfury platform's main asset, user trading data, is digitized and published using a smart contract and IPFS, a content-addressable storage system:
1. Once a position is closed the anonymized trade identifier together with transaction data including the bid and ask price, last price and timestamp are stored in an encrypted form in IPFS;
2. A hash of that data is stored on the Ethereum mainnet;
3. Quantfury publishes the key to decrypt that data on a time-delayed (30 days) basis for auditing purposes. The key decrypts the users’ monthly closed positions trading data and is saved into the Quantfury Ethereum smart contract;
4. This key can be seen via the Quantfury DAPP or by executing the Quantfury Smart Contract function to receive it.
Monetizing Trading Data
The Quantfury-developed AI employs multiple quantitative trading strategies driven by machine learning models trained on user trading data to generate income. Such a model is not practical with traditional brokerages where the trading data accuracy is negatively biased due to adjusted spreads, fees, commissions, lack of cap on sizes of user deposits, and user margin requirements for obtaining leverage. The Quantfury ecosystem is driven by the need to create the perfect trading data that is the least biased and represents the “purest” market sentiment of traders. This ecosystem is achieved by having users trade with real-time exchange buy and sell prices, without commissions and fees, and with constant trading power.
The majority of retail trader decisions are driven by the psychology of gains and losses as described in Prospect Theory (3). Trader responses to market fluctuations can be formulated into quantitative behavioral patterns based on the trader's reactions to his/her trading profits or losses.
It has been common practice for a long time that individual trading data has been monetized by financial institutions, by acquiring or owning (6) significant trade order flows that enable them to formulate profitable trading strategies. Retail traders pay commissions and various fees in order to trade, never getting a share of the monetization of trading data they generate.
Financial institutions analyze client trade order flows in a one-dimensional way, without having the full depth of the client’s trading behavior behind their trade orders. This behavior is largely determined by the size of the client’s accounts, the size of leverage they use, and their previous trading history.
The Quantfury ecosystem is designed to build a textbook-case dataset for its user trading data by:
Grouping the “fear and greed” factor of users effectively by offering constant trading power regardless of trade results;
Fixed number of initial account funding amounts at the start of trading with Quantfury, thus making a number of trader groups with similar fear and greed ratios;
Capturing the least biased or “purest” market sentiment of traders by having users trade with the real-time exchange buy and sell prices, without any commissions and fees; and
Maintaining the full depth and history of user trading decisions.
Quantfury constantly updates and applies machine learning of the trading data generated within its ecosystem. These machine learning algorithms produce automated trading decisions to buy, sell, or do nothing for each of the Quantfury incoming user trade orders. The machine learning of trader behavior, both as individuals and as groups, is conducted at each moment of the current state of the Quantfury ecosystem in real-time, thus constantly improving the financial well-being of the platform.
To maintain the efficiency and integrity of Quantfury’s machine learning algorithms and its automated trading decisions, it is a fundamental requirement to provide unmatched trading conditions to the trading app users, which in turn promises the most unbiased data of market sentiment by retail traders. Further, the platform’s predictive powers strengthen continuously as a result of allowing platform activity to occur in this entirely unbiased way.
ML Algorithms used at Quantfury:
A. Multilayer Perceptron (MLP), is a fairly simple form of artificial neural network known as a Feedforward Neural Network (FNN). FNNs are so called because information flows through them in one direction only. MLPs consist of an input layer, one or more hidden layers, and an output layer. Each neuron within the MLP performs a weighted sum of every output from the previous layer (these neurons are therefore known as densely connected), adds an additional number called a bias, and applies a non-linear function to produce its output. The non-linearity is important as it technically allows the MLP to approximate any function.
A simple 4-layer MLP is shown below:
MLPs learn (are trained), in a similar fashion to most other artificial neural networks - with the backpropagation algorithm. This algorithm is a form of supervised learning where an input is allowed to propagate through the network and then the output is compared to the ideal output. The difference between that particular output and the ideal output is known as the error. The algorithm determines which neurons, layer-by-layer, have contributed to the error and by how much and then iteratively adjusts them per example.
An autoencoder is a neural network with the purpose of distilling the important aspects of input data into a shortened form, or code. This shortened form is both useful insofar that it is efficient with respect to data size, but also because the autoencoder removes the least relevant information from the input. This allows for other machine learning algorithms to learn more efficiently if its inputs are first autoencoded.
An autoencoder usually takes the form of an MLP with hidden layers decreasing in size to a point, and then increasing back to the original size with the same number of layers thereafter. The algorithm is then trained such that it attempts to recreate the input at the output. This may seem trivial, but as the size of the hidden layers get progressively reduced, the algorithm must learn to reconstruct the input from missing information. The training process therefore intuitively learns to conserve only the most important aspects of the data.
The first part of the network is known as the encoder, while the second part is known as the decoder. After the training process is complete, only on the encoder is used in order to transform the data. A diagram outlining these parts of the network is shown below:
C. Q-learning encoder
Q-learning is part of a subset of machine learning techniques known as reinforcement learning. Reinforcement learning concerns itself with optimizing how an artificial agent takes actions within its environment in order to maximize some notion of reward in the long-term - much like how one might train an animal.
In Q-learning, there exists a set of states (S), and a set of actions (A) that may be taken at every possible state. By performing one of these actions in a particular state, the agent transitions from one state to another and earns a particular reward depending on the action and state.
The goal then, of the Q-learning algorithm is to find a function Q that is able to tell the agent what the best action is at any given state. This function is found through a training process where the algorithm is given a high-level objective that it tries to maximize - a cumulative reward from all of its actions. Examples of this include total return on investment for a trading bot, or high score for a bot attempting to learn how to play ‘pong’. The Q function can take many forms, and is commonly an MLP as discussed previously.
The learning process of a Q-learner can be described mathematically as follows:
It can be seen that the function is updated to consider both the reward the agent can earn from a current action, but also how that action might affect the ability to perform future actions which also have corresponding rewards.
Objective and Description of Quantfury’s ML Architecture:
The objective of the Quantfury ML architecture is to maximise risk–adjusted returns. By monitoring the activity of traders across the Quantfury platform, as well as integrating technical indicators from all of our instruments, we are able to construct a comprehensive view of the market state, from which our Q-learner can make profitable trading decisions. It is this system that allows Quantfury to remain profitable while offering users a fair trading environment and low margins.
B. Data Streams
Quantfury incorporates several different data streams into its ML architecture in order to accurately assess its universe. The data streams can broadly be divided into trader data and financial instrument data.
Trader data is comprised of the behaviour of traders on the platform. This data is anonymised and is simply used to model actions of agents on the platform, not track individuals. Trader data consists of a full trade history, including ROI per position, long or short, exposure, etc. This data is then further processed into rolling time windows and rolling position windows - giving an idea of how a trader’s actions have evolved per position opened and time elapsed.
The other datastream contains collated information about all the financial instruments being traded on Quantfury. This includes a large number of technical indicators such as moving averages and volatility and how these indicators have evolved over time for that particular instrument.
C. Quantfury’s AI architecture is comprised of 3 main elements - a system of deep autoencoders, a position prediction network, and a system of Q-learners.
The autoencoders are trained to intelligently extract the most vital information contained in the datastreams above. A different autoencoder is needed for each datastream as every datastream has different characteristics that must be encoded. Not only does this allow for improved performance, the autoencoders allow for compression of data, relaxing bandwidth restrictions, and improving system latency - vital for accurate and timely trades.
After all the data streams are autoencoded, an MLP makes a prediction on the performance of a particular trader’s current position, taking all current and historical information into account. This is performed for every position opened for every single instrument and a separate model.
The prediction described above is then appended to all of the autoencoded datastreams at the current timestep. This creates a state that fully describes the Quantfury universe. Such a state may then be used by the Q-learners to dictate an action that Quantfury should take - in this case, what position to take on a particular instrument and how large of a position to take. A Q-learner is trained for each instrument such that it can learn to deal with the specific dynamics of that particular instrument.
In this way, all of the models work together in order to create a real-time autonomous list of decisions that Quantfury should take at any given moment, taking into account the entire platform and overall systemic risk.
Quantfury Advantage Matrix
The Quantfury trading app was made to be the platform of choice for cryptocurrency and traditional markets trading, to offer an undeniable value proposition for crypto traders and holders, as well as traditional brokerage clients.
The following core values guide us daily to create and support a truly unique platform for millions of traders and investors who need our solution. Our values have been key for daily inspiration, making us innovate and driving us to develop the next generation product that few thought possible.
OUR USERS & PRODUCT ARE OUR OBSESSION
We are the biggest users of our product. We are the biggest proponents and critics of our product. We pay attention to competition but obsess over our product and our users.
CHALLENGE THE STATUS QUO
We want our team members to have the courage to challenge the status quo and other team members without fear. However, this should be based on logic and evidence that can be validated and defended using research rather than personal bias.
INNOVATE, CREATE AND IMPROVE
Do not listen to the experts who tell you it cannot be done. Instead, challenge the status quo if it does not make sense to you. Do not reinvent the wheel, when a simple and proven solution will do. Optimize, using cost/benefit principles, to maximize effort on the highest value tasks/deliverables.
DRIVEN BY ACTION
The speed of execution and efficiency matters in business. What can be done today we do not leave for tomorrow. We appreciate and welcome calculated risk-taking and speed of action.
COMPETE THE RIGHT WAY
Create your own ideas, work to compete, and challenge others to bring out the best out of our product and people.
How can we check the integrity of prices of executed trades in Quantfury?
Anybody can retroactively check the prices of executed trades by Quantfury trading platform users have made during the previous month by accessing https://qtf.quantfury.com. Thus, anybody can verify the correct value of QDT tokens and the number of QDT tokens generated.
What is user trade?
It is a basic economic concept for short-term gains involving a buying or selling of equities, cryptocurrencies, fiat pairs, and futures.
What is spread?
It is the difference between the bid and the ask price quoted for a specific asset type instrument, available to buy or sell.
What is liquidity?
It is the capital required for trades execution and risk management.
What is market-making?
It is an activity of constantly quoting bid and offer prices of financial instruments, always ready to buy and sell at quoted prices to fulfill the execution of the incoming trade orders.
Notes and Further Reading
Brad M. Barber, Yi-Tsung Lee, Yu-Jane Liu, and Terrance Odean. “Just How Much Do Individual Investors Lose by Trading?”. April, 2008. URL: http://faculty.haas.berkeley.edu/odean/papers%20current%20versions/justhowmuchdoindividualinvestorslose_rfs_2009.pdf
Brad M. Barber, Yi-Tsung Lee, Yu-Jane Liu, Terrance Odean, and Ke Zhang. “Do Day Traders Rationally Learn About Their Ability?”. October, 2017. URL: https://faculty.haas.berkeley.edu/odean/papers/Day%20Traders/Day%20Trading%20and%20Learning%20110217.pdf
Daniel Kahneman and Amos Tversky. “An Analysis of Decision Under Risk”.Econometrica, Vol. 47, No. 2, pp. 263-291. March, 1979. URL: https://www.uzh.ch/cmsssl/suz/dam/jcr:00000000-64a0-5b1c-0000-00003b7ec704/10.05-kahneman-tversky-79.pdf
Yang-Yu Liu, Jose C. Nacher, Tomoshiro Ochiai, Mauro Martino, and Yaniv Altshuler. “Prospect Theory for Online Financial Trading”. October 15, 2014. URL: https://dspace.mit.edu/handle/1721.1/92479
Investopedia.“National Best Bid and Offer (NBBO)”. Accessed October 15th, 2018. URL: https://www.investopedia.com/terms/n/nbbo.asp
Robinhood Financial. “SEC Rule 606 Report Disclosure, Third Quarter 2018”. Accessed October 20th, 2018. URL: https://d2ue93q3u507c2.cloudfront.net/assets/robinhood/legal/RHF%20PFO%20Disclosure.pdf