How our Artificial Intelligence uses Crowd Wisdom data to predict the crash and out-perform the market.
It’s been a tough week for retail investors around the world. Following the Coronavirus outbreak making official ‘pandemic status’ — the markets tumbled into a sell off not seen for many years.
In a global economy that is fuelled by low interest rates and cheap money — it’s not hard to see the bubbles created across asset classes. In global stocks for example, price earnings ratios are increasingly losing their meaning as a yard stick for assessing value — with bull runs on tech companies such as Tesla defying any common sense.
Unfortunately, it looks like this is a new reality. We aren’t fans of managerial acronyms here at Team Pynk — but the world of finance and investing is becoming increasingly VUCA (volatile, uncertain, complex and ambiguous).
It’s for that reason that we believe working together as one is the best way to return the odds to retail investors.
Last month, we reported in our community the increasing performance of ROSE AI and its ability to predict future pricing.
The real test for ROSE AI and the Pynk Crowd Wisdom system though has been the last week and crucially did we call the market out ahead of time? I’m happy to report that it did and below you can see the consistent performance of ROSE AI over the last 9 months.
While we collect data across many asset classes, here we are looking at performance on Bitcoin (BTC) since we have been collecting this data the longest.
Our quant team use Cumulative Log Returns as it’s easier to add over time and, being linear, represents performance over time more accurately.
For the more mathematically inclined readers — you can see more information on this here.
We take a systematic approach to testing machine learning algorithms including Linear Regression, Random Forest, Lasso, Support Vector Machine (SVM) and deep learning with Artificial Neural Networks in order to detect and select Super Predictors in the Crowd.
There is nothing particularly special about these algorithms, more so it is the way we leverage them against our proprietary data set. If the equity curve of any given algorithm against a given asset class and time frame shows the expected signs of development — we continue testing. Crucially, we must see significant outperformance of the underlying asset class and the benchmark (being the average prediction of the Pynk Crowd).
The framework which embeds both a back-test engine and a machine learning factory has been designed to seamlessly integrate new structures of Crowd data for existing users, assets and metrics, as well as to adapt to any evolution in our global Crowd of predictors.
ROSE AI plugs into state of the art open source libraries such as scikit-learn, pandas, numpy and tensorflow. This helps accelerate training times of each algorithm.
Crucially, the algorithms are structured to learn from others’ past predictions to improve users’ new predictions (beyond their knowledge).
Clusters of users are created on a continuous basis to extract and emphasise multiple features from the Crowd as a whole. We are unearthing correlations in characteristics of Crowd members’ psychometric profiles such as reactions to volatility. All the time ROSE AI and the Crowd Wisdom system is continuously exploring and validating all that academic Crowd Wisdom theory teaches us.
Academic studies on Crowd Wisdom conclude that reducing bias and noise is crucial to success. Barbara Mellers, a Wharton marketing professor and Penn Integrates Knowledge (PIK) professor at the University of Pennsylvania, and Ville Satopӓӓ, assistant professor of technology and operations management at INSEAD, examined these forces and found that noise was a much bigger factor than expected in the accuracy of predictions.¹
ROSE AI addresses these challenges in two ways. Firstly, machine learning is applied at the user level, with the expected return being especially useful. Secondly, is the ability to select the top X percent of predictors, which by definition display less bias. Finally, our approach to reducing noise is one of averaging by using a systematic machine learning approach for aggregation.
In summary, machine learning techniques are applied at several levels of the process. At the users’ level, it enhances user predictions and reduces bias, at the Crowd’s level we continuously create clusters, thus exploiting all potential features of Crowd Wisdom to make a final prediction.
IMPROVING DATA QUALITY AND QUANTITY
The really good news for Pynksters the world over is that ROSE AI will only get stronger and stronger. There are 2 key ingredients that will make that happen, quality and quantity.
Clearly as more Pynkster price predictors join the Crowd (we are now 16,700+ across 167 countries) the quantity of daily predictions and long term forecasts will continue to increase, improving ROSE AI’s ability to learn.
What is less well understood is the quality of prediction data is also increasing over time thanks to an ever increasing number of super predictors. “Superforecasters walk among us — people who can predict the future with rare accuracy, outstripping even domain experts.” ²
The below chart shows the increasing number of Super-Predictors over time (note: I have deliberately blocked the numbers since sensitive, longer lines equals more SP’s over time, and steeper lines means better accuracy);
The data tells us Pynksters are learning how to become Super-Predictors ‘by doing’, and in a risk free environment. It costs nothing to make a price prediction — in fact regular predictors even earn free shares in the Pynk investment portfolio every quarter.
Since the ability of ROSE AI and our machine learning algorithms is a function of quality and quantity — the potential for improved performance is exponential (let’s call it q squared!).
If you’re looking to join our now global movement, enjoy a more equitable model for investing, access to our 0% Fee Crowd Powered Portfolio and meet some like minded Pynksters — then why not help power our now global movement by making your first price prediction here.
If you’re an Angel Investor then check out our Seedrs Crowd Funding campaign here.
Should you have any comments or questions, please leave them below and I will personally respond as quickly as I can.
Mark
Disclaimer: Please bear in mind that this information does not constitute any form of advice or recommendation by Pynk One Ltd. and is not intended to be relied upon by users in making (or refraining from making) any investment decisions. Appropriate independent advice should be obtained before making any such decision. When investing, your capital is at risk and you may recover less than the initial investment.
- Mellers, B. Satopää, V.(2019) Want Better Forecasting? Silence the Noise. https://knowledge.wharton.upenn.edu/article/want-better-forecasting-silence-the-noise/
- Satopää, V.(2019) The Secret Ingredients of ‘Superforecasting’. https://knowledge.insead.edu/strategy/the-secret-ingredients-of-superforecasting-12721
