To the untrained eye, investment management is simply buying low and selling high. This is why so many investors try to do it on their own — and fail. On the other hand, investment management has developed into a global specialist profession. How do they do it?
The answer lies in portfolio management theory with its emphasis on risk-adjusted returns & security and inter-asset class correlations. This information is statistical in nature & requires one to trove through large data sets and make sense of them.
Essentially, what you need is a super fast computer that can look at a data set and give you the desired results. While these super-fast computers are available today and will become more prevalent in the future with the leaps in quantum computing, the problem is telling these high-speed supercomputers what to look for.
Sure you can write code for it but what if you want to now remove factor “A” and add factor “B”? Do you write code for it again?
And what about the instances where a result that you’re looking for leads to more questions? That is how the human mind thinks where answers lead to more questions. How does a one-time code snippet handle that?
More importantly, why must you tell the supercomputer to do something? Can’t it be trained to do it on its own?
Machine Learning — The Future?
This is where Machine Learning comes in as the branch within artificial intelligence concerned with statistical algorithms and how code can be written to think like humans and perform tasks without the need for constant / repeated instructions. The main focus of these models is to develop predictive power by making inferences on the underlying data set and then shaping and molding the data for the best optimization of results.
Feeling a little lost? No problem. An example will help clear it up.
Consider a 3-year data set comprising trader returns & the time they take to make redemptions. Data mining would take the population, generate a sample, and tell us that X number of traders take < 5 days to make redemptions whereas Y number of traders take > 5 days. A subsequent analysis would reveal that of those who take > 5 days, 70% are those with negative returns.
A machine learning approach would be to analyze the entire data for patterns & suggest that delays in redemptions are directly correlated with the return profile of the trader, that is, a trader with negative returns would be less inclined to close a position at a loss and meet the redemption call as compared to a trader with positive returns who would be more comfortable in doing so.
The machine, in this instance, could be trained to find other correlations as well, including whether the traders with negative returns are those who use high leverage or compromise on position sizing and stop loss. Thus, the underlying data shifts and molds till we have our conclusion that traders who have a high-risk approach are bound to be poor at meeting redemptions.
Zignaly & Machine Learning
As can be inferred, machine learning is a far more powerful approach that yields greater insights into the data set. More importantly, it is far quicker than a normal data mining approach. This is especially important from the perspective of the fast-paced world of trading with its high volatility needing a high-touch approach.
At Zignaly, our Z-Score algorithm, powered by artificial intelligence, uses these deep-rooted concepts of machine learning on a 5-year (and expanding) data set of traders and their different metrics to rank the best amongst them. This ensures that investors on the Zignaly platform have access to only the best traders and that they can have comprehensible tools to be able to differentiate between traders.
With our expanding focus on DeFi, the universe of all possible investment opportunities that investors could co-invest in with expert traders is set to increase exponentially. So as our data set expands with time & focus, a simple data mining approach would continue to become less optimal from the perspective of economizing on the time new information reaches our investors. With machine learning, our Z-Score continues to be an autonomous agent that is continually building & relaying predictive power through pattern analysis.
You may argue that patterns may throw up spurious correlations as well — like trader returns being negative on every alternate Sunday of the month. However, this is more a problem with sample size in data mining and less with machine learning as the latter uses the complete data set, instead of a sample, and can be trained to judge the strength of the pattern as well.
Stakeholder Objectives — Goal Congruence
The goal congruence here, as a performance-first social investment platform, is high between traders, investors & $ZIG holders:
- Investors get timely information with predictive power to make the best decisions on which trader to invest with
- Traders with high Z Scores attract more capital and continue to churn out profitability for themselves but also for the platform and the investors.
- The platform revenue increases enabling more availability for the fortnightly buyback and burn program. This makes $ZIG a deflationary asset.
We hope this has been helpful in aiding your comprehension of Z-Score and how it creates a network effect that leads to the best outcomes for all stakeholders. Stay tuned for more as we continue our coverage of the AI-powered revolution.
