Forecasting of Volatile Assets using Artificial Swarm Intelligence

Abstract

Swarm Intelligence (SI) is a natural process that has been shown to amplify decision-making accuracy in many social species, from schools of fish to swarms of bees. Artificial Swarm Intelligence (ASI) is a technology that enables similar benefits in networked human groups. The present research tests whether ASI enables human groups to reach more accurate financial forecasts. Specifically, a group of MBA candidates at Cambridge University was tasked with forecasting the three-day price change of 12 highly volatile assets, a majority of which were cult (or meme) stocks. Over a period of 9 weeks, human forecasters who averaged +0.96% ROI as individuals amplified their ROI to +2.3% when predicting together in artificial swarms (p=0.128). Further, a $5,000 bankroll was managed by investing in the top three buy recommendations produced each week by ASI, which yielded a 2.0% ROI over the course of the 9-week study. This suggests that swarm-based forecasting has the potential to boost the performance of financial traders in real-world settings.

Full text

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE
Forecasting of Volatile Assets using Artificial Swarm
Intelligence
Louis Rosenberg
Unanimous AI
San Francisco, USA
louis@unanimous.ai
Gregg Willcox
Unanimous AI
San Francisco, USA
gregg@unanimous.ai
Martti Palosuo
Judge Business School
University of Cambridge
Cambridge, UK
prm50@jbs.cam.ac.uk
Ganesh Mani
Carnegie Mellon University
Pittsburgh, USA
ganeshm@andrew.cmu.edu
Abstract— Swarm Intelligence (SI) is a natural process that has
been shown to amplify decision-making accuracy in many social
species, from schools of fish to swarms of bees. Artificial Swarm
Intelligence (ASI) is a technology that enables similar benefits in
networked human groups. The present research tests whether ASI
enables human groups to reach more accurate financial forecasts.
Specifically, a group of MBA candidates at Cambridge University
was tasked with forecasting the three-day price change of 12
highly volatile assets, a majority of which were cult (or meme)
stocks. Over a period of 9 weeks, human forecasters who averaged
+0.96% ROI as individuals amplified their ROI to +2.3% when
predicting together in artificial swarms (p=0.128). Further, a
$5,000 bankroll was managed by investing in the top three buy
recommendations produced each week by ASI, which yielded a
2.0% ROI over the course of the 9-week study. This suggests that
swarm-based forecasting has the potential to boost the
performance of financial traders in real-world settings.
Keywords—Artificial Swarm Intelligence, Swarm Intelligence,
Human Forecasting, Financial Forecasting, Investing, Group
Forecasting, Cult Stocks, Meme Stocks, Collective Intelligence,
Wisdom of Crowds, Human-Machine Teaming.
I. INTRODUCTION
It is well known that groups of forecasters can outperform
individuals by aggregating estimates using statistical methods
[1-3]. Often called the Wisdom of Crowds (WoC) or Collective
Intelligence (CI), this phenomenon was first observed over a
century ago and has been applied to many fields. The most
common methods involve polling human groups and then
aggregating their input as a simple or weighted mean [4].
Recently, a new method has been developed that is not based
on aggregating input from isolated individuals but involves
synchronous groups of forecasters working together as real-time
systems. Known as Artificial Swarm Intelligence (ASI) or
Swarm AI, this method has been shown in numerous studies to
significantly increase the accuracy of group forecasts [5-13].
In a recent study at the Stanford University School of
Medicine, groups of doctors were asked to review chest X-rays
and predict the likelihood that each patient had pneumonia.
When working together in artificial swarm, diagnostic errors
were reduced by over 30% [14]. In another study, groups of
financial traders were asked to predict common market
indicators including the price of gold, oil, and the S&P 500.
Results showed a 36% increase in forecasting accuracy when
participants used ASI as compared to traditional methods [20].
While prior studies have shown ASI to significantly amplify
the group accuracy in controlled settings, the present work
assesses whether swarm-based forecasting of highly volatile
assets (mostly so-called cult or meme stocks), achieves similar
improvements. To address this, a nine-week pilot study tasked a
group of MBA candidates at Cambridge University with making
weekly forecasts of 12 high-volatility assets, comparing
individual forecasts to swarm-based predictions. Performance
was also compared to traditional Wisdom of Crowd methods.
II. BACKGROUND
A. Swarm Intelligence (SI)
The decision-making process that governs honeybee swarms
has been researched since the 1950s and has been shown at a
high level to be quite similar to decision-making in neurological
brains [15,16]. Both employ populations of simple excitable
units (i.e., neurons and bees) that work in parallel to integrate
noisy evidence, weigh competing alternatives, and converge on
decisions in real-time. In both brains and swarms, outcomes are
arrived at through competition among groups of excitable units.
In honeybees, this enables hundreds of scout bees to collect
information about their local environment and then deliberate in
synchrony, converging on a single optimal decision [17-20].
In the natural world, swarming organisms establish real-time
feedback loops among group members. To achieve this among
groups of networked humans, ASI technology allows distributed
users to form closed-loop systems moderated by swarming
algorithms [5-9]. The goal is to enable groups of distributed
users to work in parallel to (a) integrate noisy evidence, (b)
weigh competing alternatives, and (c) converge on decisions in
synchrony, while also allowing all participants to perceive and
react to the changing system in real-time, thereby closing a
feedback loop around the full population of participants.
B. Swarm Software
The software used to enable ASI in this study is called the
Swarm® platform from Unanimous AI and is shown in Figure 1.
Using this software, groups answer questions in real time by
collaboratively moving a graphical puck to select among a set of
answer options. Each participant provides input by moving a
graphical magnet to pull on the puck, thereby imparting their

personal intent on the system as a whole. The input from each
user is not a discrete vote, but a stream of time-varying vectors.
Fig. 1. Users engaging Swarm software to rank assets.
Because all users can adjust their intent continuously in real-
time, the puck moves based on interactions among all members,
empowering the group to converge in synchrony. Participants
must continuously update their input throughout the real-time
process or lose their influence over the outcome. This enables
the intelligence algorithms to continuously monitor the changing
behaviors of all members, modulating the aggregation. Figure 2
shows an example of the underlying human behaviors. More
details on the Swarm software can be found in [21, 22].
Fig. 2. Behavioral plot of the real-time decision making process. Darker
areas convey higher conviction. The dotted line shows the puck trajectory.
III. METHODOLOGY
To assess the ability of human groups to forecast cult stocks,
we conducted a nine-week study using volunteers from the
Cambridge Judge School MBA program. Volunteers self-
identified as interested in cult stocks and followed at least one
stock closely. In other words, they were all representative of the
high-level demographic driving the cult-stock movement. Each
weekly group of between 8 and 16 participants came from the
same pool of volunteers. To establish a baseline, all participants
provided their weekly forecasts as individuals using a standard
online survey. The group then congregated online in real-time
and used the Swarm platform and make collective forecasts.
In each week of the study, participants first predicted the
price change of the 12 assets over the next three days in a survey.
The survey asked participants to buy or short up to 2 units of
their virtual bankroll for each asset, and to predict which asset
would increase the most and which would decrease the most
respectively. The Wisdom of the Crowd (WoC) response to each
question was calculated as the most popular response provided
by the survey participants (i.e. the statistical Mode).
Next, participants logged into the Swarm platform to
synchronously answer these same questions as a group—first
allocating their virtual bankroll for each stock, and then creating
two rankings of assets: the most likely to decrease and the most
likely to increase over the next three days. For these rankings,
the top 5 longs and top 5 shorts were considered. All individuals
were anonymous to one another while swarming.
Swarm sessions started approximately 15 minutes after the
close of the market and lasted approximately half an hour. The
price of each security was recorded at the open of the market the
day after the swarm, and also at the close of the market three
days after the swarm. The price of bitcoin (BTC) was recorded
as soon as the swarm ended, as BTC trades continuously.
The percentage price change in each stock was calculated
using the price of the equity upon market open the day after the
swarm and the price at close of market three days after the
swarm. The top three individuals whose virtual bankroll showed
the highest ROI over the three-day period were awarded
bonuses: $15 for first place, $10 for second place, and $5 for
third place. This bonus was to incentivize participants to use
their best efforts in the forecasting surveys.
Finally, we managed a real bankroll over the duration of this
experiment: starting with $5,000, we invested in the three stocks
the Swarm ranked as most likely to increase in price. Bonuses
were awarded to participants based on the overall performance
of the swarm-managed bankroll. This was done to motivate best
efforts from members during the swarming portion of the study.
A. Data Analysis
Of the 108 asset movements collected, the mean movement
was 1.62% upwards, which was skewed higher by the presence
of a small number of outliers that increased in price by more than
20%. No price ever decreased by more than 20%.
Fig. 3. Distribution of Observed Price Movements

Such extreme outliers distort the analysis and interpretation
of these results by biasing towards a handful of data and may be
unrealistic in practice: a trader or hedge fund manager would
likely reduce exposure to these wild events by using stop losses
(and perhaps profit targets). As a result, we consider a clipping
function that restricts the maximum movement of these stocks
in the three-day window to a fixed interval: either 10% or 20%.
For reference, the 12 assets forecast consisted of stocks
(GME, AMC, RKT, TLRY, PLTR, TSLA, PTON, AAPL and
SPCE), ETFs (ARKK and social-media driven BUZZ) and
Bitcoin (BTC, a volatile cryptocurrency). A chart of the average
price change of each asset is provided in Figure 4. The vast
majority of price movements were under 5% during each trading
period, though some assets exhibited larger volatility.
Fig. 4. Average Movement across 12 Assets with 95% Confidence Interval
shown as black bars.
Finally, to meaningfully compare data, we ran statistical
significance tests comparing each investment strategy to each
other investment strategy by bootstrapping 1,000 times over the
data points produced by each metric. For example, to bootstrap
the Swarm Top Picks metric over the 9 weeks of the study, we
randomly selected 18 of the swarm’s Top Long and Top Short
picks with replacement from the set of 18 data points (9 weeks,
one Long and one Short per week) we have for this metric.
B. Results
When using a 20% clipping function, the swarm’s top-
ranked picks netted an average ROI of 2.3% week-over-week,
which outperformed the Individual (0.96%, p=0.13) and WoC
(1.6%, p=0.18) top-ranked picks, as shown in Figure 5. As a
result, we can be more than 80% confident that the top-ranked
Swarm picks outperformed the WoC and Individual rankings
due to more than random chance. Over the 9-week course of this
study, the Swarm Top Ranking strategy’s weekly average ROI
corresponds to an estimated cumulative ROI of 22.7%.
Fig. 5. Investment Performance using 20% Clipping.
To examine how sensitive these results are to changes in the
clipping, we next limited the stock movements further using a
10% clipping function. In this context, the swarm achieves a
1.77% ROI, which outperforms the average individual (-0.26%
ROI, p=0.110) and the WoC (-0.39% ROI, p=0.106). As a result,
we can be more than 85% confident that the Swarm Rankings
outperformed both the average individual and the median
individual response in this respect due to more than random
chance alone. We also see that there’s a reasonable range of
clipping limits for which the Swarm Top Rankings outperform
the WoC and Individual top rankings.
Fig. 6. Investment Performance using 10% Clipping.
We also find that the Swarm Top Rankings achieved a
significantly positive ROI in both the 20% clipping condition
(p<0.1) and the 10% clipping condition (p<0.1). Neither the
WoC nor Individual Top Rankings for either of these clipping
conditions yielded a significantly positive ROI (p>0.25 in all
cases). The distribution of all 1000 bootstrapped performances
for each investment strategy is shown in Figure 7.

Fig. 7. Distribution of Bootstrapped Performance using 10% Clipping
To more accurately evaluate the quality of these strategies
compared to the market, we can examine the performance of a
benchmark index, the S&P 500 (SPY), over the same time
intervals. We tabulated the price changes of SPY across the
length of this study in the same way as the cult stocks. SPY on
an average increased by +0.4% over the same time period we
considered. The Swarm Top Rankings outperformed this
benchmark metric using both the 10% clipping (p<0.2) and 20%
clipping (p<0.2), while the Individual and WoC Top Rankings
did not. As a result, we can conclude that Artificial Swarm
Intelligence allowed this group to achieve a higher ROI in
forecasting cult stocks than they would have achieved by
investing in a market benchmark.
C. Real Bankroll Analysis
Before the experiment began, we allocated a real bankroll of
$5,000 and invested weekly by splitting the full bankroll across
the top three swarm-ranked buy picks for that week. At the end
of the nine-week experiment, this bankroll had grown by 2.0%.
IV. CONCLUSIONS
This study highlights a promising technology for amplifying
the real-world forecasting power of groups: Artificial Swarm
Intelligence (ASI). In this study, ASI enabled a group of MBA
candidates to forecast the price movements of 12 high-volatility
financial assets (chiefly, equities) more accurately than if the
group were forecasting as individuals or as a crowd aggregating
their input statistically. The swarm-based forecasts yielded an
impressive 22.7% cumulative ROI over the nine-week study by
selecting one “top long” and one “top short” from the 12 assets
under consideration each week. Further, the top three long picks
were used each week to manage a real-world bankroll (with
fees) and achieved +2.0% ROI over the nine-week test.
These results add to previous research demonstrating that
human groups can use Swarm AI to make better collective
assessments across a wide range of domains, from subjective
judgements and medical diagnoses to market forecasting [5-14].
While this study was limited in that it only involved forecasting
volatile cult stocks with groups of MBA students, the results
support prior research showing success amplifying the accuracy
of group financial forecasts using Swarm AI [12].
ACKNOWLEDGMENT
The authors thank the MBA candidates at the University of
Cambridge, for their time. Thanks also to Culley Deisinger and
Chris Hornbostel for moderating and organizing the sessions.
REFERENCES
[1] Galton, F. (1907). Vox Populi. Nature, 75, 450-451.
[2] Steyvers, M., Lee, M.D., Miller, B., & Hemmer, P. (2009). The Wisdom
of Crowds in the Recollection of Order Information. NIPS. 1785-93.
[3] Philip E. Tetlock and Dan Gardner. 2015. Superforecasting: The Art and
Science of Prediction. Crown Publishing Group, New York, NY, USA.
[4] J Dana, P Atanasov, P Tetlock, B Mellers (2019), Are markets more
accurate than polls. The surprising informational value of “just asking.”
Judgment and Decision Making 14 (2), 135-147.
[5] Rosenberg, L.B., “Human Swarms, a real-time method for collective
intelligence.” Proceedings of the European Conference on Artificial Life
2015, pp. 658-659.
[6] Metcalf L, Askay DA, Rosenberg LB. Keeping Humans in the Loop:
Pooling Knowledge through Artificial Swarm Intelligence to Improve
Business Decision Making. California Management Review.
2019;61(4):84-109.
[7] Patel, B.N., Rosenberg, L., Willcox, G. et al. Human–machine partnership
with artificial intelligence for chest radiograph diagnosis. npj Digital
Medicine 2, 111 (2019). https://doi.org/10.1038/s41746-019-0189-7
[8] Rosenberg, Louis. (2016) “Artificial Swarm Intelligence vs Human
Experts,” International Joint Conference on Neural Networks (IJCNN).
[9] Rosenberg, Louis. Baltaxe, David and Pescetelli, Nicollo. "Crowds vs
Swarms, a Comparison of Intelligence," IEEE 2016 Swarm/Human
Blended Intelligence (SHBI), Cleveland, OH, 2016, pp. 1-4.
[10] Baltaxe, David, Rosenberg, Louis and N. Pescetelli, “Amplifying
Prediction Accuracy using Human Swarms”, Collective Intelligence
2017. New York, NY.
[11] Willcox G., Rosenberg L., Askay D., Metcalf L., Harris E., Domnauer C.
(2020) Artificial Swarming Shown to Amplify Accuracy of Group
Decisions in Subjective Judgment Tasks. In: Arai K., Bhatia R. (eds)
Advances in Information and Communication. FICC 2019. Lecture Notes
in Networks and Systems, vol 70. Springer, Cham.
[12] L. Rosenberg, N. Pescetelli and G. Willcox, "Artificial Swarm
Intelligence amplifies accuracy when predicting financial markets," 2017
IEEE 8th Annual Ubiquitous Computing, Electronics and Mobile
Communication Conference (UEMCON), NY, NY, 2017, pp. 58-62.
[13] L. Rosenberg and G. Willcox, "Artificial Swarm Intelligence vs Vegas
Betting Markets," 2018 11th International Conference on Developments
in eSystems Engineering (DeSE), Cambridge, UK, 2018, pp. 36-39.
[14] L. Rosenberg, M. Lungren, S. Halabi, G. Willcox, D. Baltaxe and M.
Lyons, "Artificial Swarm Intelligence employed to Amplify Diagnostic
Accuracy in Radiology," 2018 IEEE 9th Annual Information Technology,
Electronics and Mobile Communication Conference (IEMCON),
Vancouver, BC, 2018, pp. 1186-1191.
[15] Seeley T.D, Buhrman S.C 2001 “Nest-site selection in honey bees: how
well do swarms implement the ‘best-of-N’ decision rule?” Behavioral
Ecology and Sociobiology. 49(5), 416–427.
[16] Marshall, James. Bogacz, Rafal. Dornhaus, Anna. Planqué, Robert.
Kovacs, Tim. Franks, Nigel. “On optimal decision-making in brains and
social insect colonies.” Soc. Interface 2009.
[17] Seeley, Thomas D., et al. "Stop signals provide cross inhibition in
collective decision-making by honeybee swarms." Science 335.6064
(2012): 108-111.
[18] Seeley, Thomas D. Honeybee Democracy. Princeton Univ. Press, 2010.
[19] Seeley, Thomas D., Visscher, P. Kirk. “Choosing a home: How the scouts
in a honey bee swarm perceive the completion of their group decision
making.” Behavioural Ecology and Sociobiology 54 (5) 511-520.
[20] Schumann, H., et. al. (2019) “‘Human Swarming” Amplifies Accuracy
and ROI when Forecasting Financial Markets. IEEE International
Conference on Humanized Computing and Communication (HCC), 77-
82.
[21] Rosenberg L., Willcox G. (2020) Artificial Swarm Intelligence. In: Bi Y.,
Bhatia R., Kapoor S. (eds) Intelligent Systems and Applications.
IntelliSys 2019. Advances in Intelligent Systems and Computing, vol
1037. Springer, Cham. https://doi.org/10.1007/978-3-030-29516-5_79
[22] ASI White Paper - https://unanimous.ai/whitepaper