Technical ReportPDF Available

DeFiPaper

Authors:
  • DeFinance Technologies Oy

Abstract and Figures

Financial assets trading using application programming interfaces (API) is increasingly becoming common for both individuals and companies. With API, it is important that the traders get their trading signals accurately in order not to lose their capital amidst the highly volatile markets. Generating buy-sell signals for newer markets such as cryptocurrency or decentralized finance (DeFi) is more complex compared to conventional markets. This release of the white paper entails a newly developed financial markets indicator (EVX) and leverages methods of artificial intelligence (pattern recognition and machine learning) to determine the buy-sell conditions.
Content may be subject to copyright.
LEVERAGING EVX: A HIGHLY
SCALABLE FINANCIAL
MARKETS INDICATOR FOR
DEFI
[Technical whitepaper – release 2– revision 1.0]
March 26,2022– DeFinance Technologies R
contents
1Executive summary 2
2Problem Description 3
2.1Seasonalityproblems............................. 3
2.2Systemic risk amplification . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3Liquidityconstraints ............................. 4
3Research and Methods 5
3.1TheWorkingofEVX ............................. 5
3.2TheMethods.................................. 7
3.3Populating Buy-Sell Signals . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.4Productionmodels .............................. 8
4Case study 8
4.1Assetsscreening................................ 8
4.2Optimized assets selection . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5Risk assessment and monitoring 10
5.1Drawdownoptimization .......................... 10
5.2Otherimportantratios ............................ 10
6The business model 12
6.1DeFinanceEconomics ............................ 12
6.2SubscriptionBasis............................... 12
7Summary 13
1
executive summary 2
abstract
Financial assets trading using application programming interfaces (API) is increas-
ingly becoming common for both individuals and companies. With API, it is im-
portant that the traders get their trading signals accurately in order not to lose their
capital amidst the highly volatile markets. Generating buy-sell signals for newer
markets such as cryptocurrency or decentralized finance (DeFi) is more complex
compared to conventional markets. This release of the white paper entails a newly
developed financial markets indicator (EVX) and leverages methods of artificial in-
telligence (pattern recognition and machine learning) to determine the buy-sell con-
ditions.
1 executive summary
Finance as a discipline has for a long time been known to be an art. Particularly
that of managing money flow streams. Other literature defines finance as a disci-
pline concerned with allocation or investment of assets over space and time, often
under the constraints of risks and uncertainties. It therefore entails the actors as
participants, aiming to price assets based on factors such as: risk levels, expected
rate of returns on investment (ROI), time value of money, etc. The last few decades
have however overseen tremendous unprecedented evolution in the history of fi-
nance, both as a discipline and the actors aforementioned. As opposed to humans
as the sole actors hitherto, standalone algorithmic trading engines have also joined
the block.
The ultimate objective in most asset trading activities is knowing when to get into
and out of a trade. Obviously, by “when” we not only imply the time aspect, but
must also take into account other factors such as the market forces of demand, sup-
ply, asset quantities, and the prices that are likely to to maximize our gains. Success
in asset trading is therefore heavily dependent on accurate timing of the trades
[1]. There are a handful of technical indicators that are well established and tested,
some of which are: simple moving average (MA), exponential MA, stochastic os-
cillator, MA convergence divergence (MACD), Bollinger brands, relative strength
index (RSI), Fibonacci retracement, Ichimoku cloud, standard deviation, and aver-
age directional index (ADX) among others. The rule of the thumb however, is that
we do not use a single indicator in isolation. In fact hardly does each one of these
indicators work for a foreseeable future when implemented inherently. Instead, it
takes the correct combination of these algorithms.
However much the above indicators have been applied successfully, there are prob-
lems associated with each one of them, resulting in a range of errors from pure
mathematical to simple dysfunctions due to market dynamics. In essence, all these
indicators boil down to two words overbought and oversold. The fundamental ob-
jective in all with no exception, is to determine the characteristics of assets before,
during, and after changing hands.
The focus of this whitepaper is to introduce an excess volume-based methodology
that attempts to estimate and compare the bids and the ask volumes, and uses them
to project the following attributes of the target market:
i The dynamic bid-ask volumes, besides the static (total) volume provided by
the exchanges and data providers
ii Determine the exact bid-ask spread, and
*DeFinance Technologies Oy, Helsinki, Finland
problem description 3
iii Use these information to estimate an averaged index quantifying all the con-
ventional indicators in one go, which is then used to generate buy-sell signals.
For instance, setting EVX to a certain index value, a combination of equiva-
lent RSI =value 1,ADX =value 2, ...Indicator N=value n are simultane-
ously set and used automatically.
The hypothesis of excess volume method is to aim and position the user into a trade
only when the asset’s supply is just enough to be available, the market demand is
there (liquidity), and the price is just right to yield a reasonable return on invest-
ment. The biggest hurdle to algorithmic trading is the markets’ unpredictability,
that leads to inconsistencies in the results produced by most systems in the long-
term.
The outgoing rationale thus justifies the need for continuous search for better algo-
rithms in order to sustain the competition with arising needs for high frequency,
yet accurate financial operational and technical analysis. We therefore introduce a
new technical indicator dubbed Excess Volume Index (EVX). At best EVX tries to
consolidate the problems associated with aforementioned earlier forms of market
indicators and address them all in one. An important point to note with the in-
dex is how it takes easily accessible data (OCHLV), and extract up to 95% accurate
bid-ask spread, not only as a percentage, but actual volumes of successful bids and
asks. It then uses this information to estimate the momentum index that can easily
be incorporated to populate buy-sell signals that are then sent as requests through
application programming interface(s) (API).
2 problem description
The problems associated with applying these old indicators can be categorized into
the following groups:
Seasonality problems
Systemic risk amplification
Liquidity constraints
2.1 Seasonality problems
Even the best algorithm today ends up mischaracterizing the buy-sell conditions
when sufficient run-time is provided. An ideal algorithm should be able to with-
stand seasonal changes resulting from stock market climate, social and political
sentiments, regardless of the extent of the time it is in use. The challenge of course
is capturing these effects and successfully corroborating them within the guiding
mathematical principles. By taking into account not only the actors, but also their
projected intentions following an event, and the rate at which they act when they
do, EVX attempts to model the gradients in which the asset prices and volumes
changes based on these factors. By doing this, the algorithm naturally attains
semi-immutable characteristics to an extent of high scalability where as little as
one month backtesting data can be used to forecast as long as one year’s expected
returns on investment. All these, while retaining the algorithm’s rigour.
problem description 4
2.2 Systemic risk amplification
Systemic risk is nowadays conventional wisdom, especially after the financial cri-
sis of 2008. By definition, systemic risk is the risk that the financial system ceases
to perform its function of allocating capital to the most productive use because of
financial difficulties among a significant number of financial institutions [2]. The
view that the architecture of the financial system plays a central role in shaping sys-
temic risk is increasingly starting to be accepted as an explanation [3]. The specific
aspect of architecture referred to, is the interconnectedness. When a one system is
more connected to other systems, it is exposed to a higher level of systemic risk[4].
But what does EVX indicator has to do with systemic risk? Well, there is a noticeable
trend between say the tanking in the US $ market and (the rise in) specific assets
in the NYSE. The same case applies to the cryptocurrency market, and is even
more severe due to the high correlation between say some top alt-coins and the
bitcoin market. There is thus a noise signal affecting mostly all known market
indicators due to this interconnectedness. A clearer signal is obtained by filtering
out these noises through direct estimation of the bid-ask volumes, and thus the
spread, together with an assessment whether the asset is overbought or oversold.
The buy signals are thus generated not only based on the market and momentum
movements, but also on deeply characterized nature of each asset encoded in the
algorithm. EVX thus leverages all the above strengths to return a superior algorithm
that maximizes gains while minimizing susceptibility to the asset’s risks such as
volatility and draw downs.
2.3 Liquidity constraints
There are two exemplary problems when designing asset trading algorithms: the
easy problem entailing definition of the variables, deciding on sufficient time frames
or ticker intervals, setting the type of order, be it limit, market, stop-limit or any of
the many order types. Then there is the hard problem: deciding when to open a
position and when to close, is not enough. You also want to make sure there is
sufficient in(out)flow of resources. This is to imply that when you open a position,
you should be able to close it whenever you want. While this is seemingly obvious,
some assets do not have sufficient liquidity, and may not exit the portfolio at the
rate it is required – what would be termed as slippage.
Liquidity is not inherently an asset property, but but a complexity of the condi-
tions in which the asset was bought, together with those in which about to be sold.
For instance, an asset acquired while it was overbought will generally have more
difficulty to dispose (even if the market shows profitability) than one acquired in
oversold conditions. This aspect of liquidity or lack of it thereof is addressed by
using EVX, as it will attempt to trade only when the odds are tilted towards the
trader’s favor. With EVX, populating the buy signals is as easy as setting the per-
cent in which the asset needs to be oversold before buying. The inverse problem is
used under oversold conditions, and when the right timing is used, more than half
of the liquidity constraints of a portfolio are addressed by the EVX.
research and methods 5
3 research and methods
3.1 The Working of EVX
The profit model describing change between the selling and buying price is the most
practical artifact we know of in the world of merchandising and entrepreneurship
today. It is simply the difference between the two prices of an asset, call the selling
price y, and bought at xboth in US $ per share. In the strictest sense the change, c
according to this model is expressed as follows:
c=|y||x|(1)
We desire that we get out of a trade when the change is positive, rather the differ-
ence between the closing and opening rates is sufficiently greater than zero. The
problem with the model in Eq.(1) is that it looks at only two points, the beginning
and the end of the transaction time frame, also known as the ticker interval. An
ideal model is the one that projects and describes the change as a fluid continuum,
aiming to capture what happens both between and within the time frame in which
the transaction occurs.
The excess volume concept also implies that the successful bids volume, Bare nec-
essarily more than asks, Aat the close of the trading time represented by the ticker
interval if any profit is to be made. As we always do, we test this hypothesis with
historical data, where we are only given in form of open, high, close, low and vol-
ume (OHCLV). A trade thus has a high expectancy when B >> A. That is the
hypothesis of excess volume methodology. But how we determine these volumes
from historical data is the focus here, as this determines the bid-ask spread and
most importantly constitutes feature extraction used to obtain the training and pre-
diction data for machine learning models.
To successfully model the intricacies between and within the ticker intervals, we
introduce an old mathematical concept referred to as perturbation theory. This means
we are seeking for (non-linear) function(s) that estimates the original independent
variable, i.e. cin this case. EVX hypothesizes that the change in price is initiated
and led by the difference between the bids and ask volumes. In other words, the
infinitesimal change can be described with differentials as follows:
δV =
N
X
i=0
(Bi±Ai)(2)
The hypothesis can thus be summarized in into the following mathematical theo-
rems:
Theorem 1.: At the start of the change, the bid-ask volume spread is the highest, and as the
change progresses, the volume difference spread tightens, zeroing in at the last millisecond
of the ticker interval.
Proof. We have that c=|y||x|. We can perturb the change such that it becomes: c=
2|y|[f(V)+|x|]. These two equations are innately the same, with the assumption
that f(V)will build up to equal exactly yat the end of the ticker interval. Then,
the change will be given simply as c=2y (y+x)same as the original form. The
rest of the exercise in the following sections is thus to estimate the function of asset
volumes, f(V)in terms of OHCLV as given by most financial data providers.
Theorem 2.: The product of the evolving change in price and the infinitesimal volume in
Eq.(2)yields the bid-ask spread.
research and methods 6
Proof. By above definition: the volume spread can be expressed in the following
manner: cδV =Bδy Aδx. As the change progresses towards the end of the ticker
interval, the differentials can now be treated as full partial derivatives as will be
seen in the next sections. The differentials δy δx =2y ynx, where ynis a
constant function of volume as described in the perturbed change.
The second derivative of volume with respect to bids, asks volumes, opening and
closing prices will be given as follows:
2V
∂B∂y =1
cand (3)
2V
∂A∂x = 1
c(4)
Combining the two equations in Eq.(4) (subtraction) gives us a full partial differen-
tial equation of the following form when the price change is introduced:
2V
∂B∂y ±2V
∂A∂x =2
c∂c
∂y +∂c
∂x (5)
The LHS of Eq.(5) opens up a new discussion with the ±symbol. What this means
is that we now have infinite solutions to the equation. The three obvious solutions
are as follows:
c=|y||x|, (6)
c=2|y|(yn+|x|)and (7)
V=B+A(8)
We have two solutions for the price change, but only one for the volume spread.
The ultimate desire is to estimate the exact values of asks, A and bids B, and track
their percentage changes giving rise to the EVX momentum indicator. Once these
values are estimated, the momentum indicator will then be calculated and used to
generate the buy-sell signals. EVX indicator therefore does the following:
i It informs us when to send a buy signal, as we desire to enter into a trade only
when bids are higher than the asks, and the spread is wide enough.
ii We also desire to exit the trade when conditions in [i] above are still valid.
Meaning, there is a high expectancy of profitability when we conduct our
trading activities within the window when B >> A.
When the change in price is presumed to take the classical form of Eq.(1), the sign
of LHS of Eq.(5) is (+), both sides of the equation reduces to zero i.e.:
2V
∂B∂y +2V
∂A∂x =2
c∂c
∂y +∂c
∂x (9)
where c=yx, and V=B+A:
∂c
∂y +∂c
∂x =0(10)
The above system is thus non-trivial. On the other hand, when the change is taken
to be the perturbed form:
2V
∂B∂y 2V
∂A∂x =2
c∂c
∂y +∂c
∂x (11)
Where c=2y (yn+x). It is seen that the RHS remains 2/c, since:
∂c
∂y +∂c
∂x =1(12)
research and methods 7
This prompts the need to establish function(s) of volume that will fully satisfy the
equation. A unique solution will effectively split the bids and asks into exact values
and always corresponds to the whole range of data on the OHLCV data as provided
by the exchanges. The methods section is designated to prove that such functions
exists and how they can be leveraged to populate the buy-sell signals.
3.2 The Methods
3.2.1 Problem formulation
Excess volume index by definition is the percentage change in the bids-ask volumes
over the time frame in which the computations are taken, also referred to as the
ticker interval. The premise of the hypothesis is that if the spread has increased by
a certain amount in the last ticker interval, the same trend is expected to continue in
the next time frame provided such time interval is sufficiently small. This implies
that the indicator is estimated as follows:
EVX =BA
A(13)
We are required to establish the exact amounts of B and A for each ticker interval
in order to estimate the momentum indicator. There are two ways to achieve this:
Analytically or numerically. We can solve either of Eq.(11) or Eq.(12) with appropri-
ate boundary conditions as set in section 3.1. In order to solve Eq.(11) analytically,
the problem is necessarily set as follows:
VBy VAx =2
c,
V(x,0)=V(x),
V(0,y)=V(y),
V(x,y)=F(y)+G(x)
(14)
Alternatively, solving Eq.(11) analytically will entail the following formulation:
cycx=1,
c(0,y)=c(x),
c(x,0)=c(y),
c(x,y)=E(y)+H(x)
(15)
3.2.2 Solution
The above formulation results in the thinking that asks, A and bids, B are ex-
pressible as pure functions of the opening and closing prices respectively following
d’Alembert’s solution where then:
V∂c
∂y +V∂c
∂x =F(y)+G(x)(16)
The price changes in Eq.(1) and (10) are thus reflected as the characteristic equations.
The solution in Eq.(16) can therefore be wrapped up into a mathematical theorem.
Theorem 3.In an exchange volume-space topology, the change in asset prices are the axes
surrounded by contours of constant (excess) volume. The gradients between the peaks and
the troughs dictates the buy-sell signals generated.
case study 8
Figure 1: Illustration of excess volume contours
3.3 Populating Buy-Sell Signals
Upon estimating the bid-ask spread from historical data, these values are then run
through a rigorous pattern recognition attempting to replicate future (un-seen) sce-
narios. Generating the buy-sell signals with EVX then become as simple as drawing
a matrix between the asset’s asks, bids, and current (open and close) prices. The
determinants of the matrix dictates whether it is worth-wile entering or exiting a
trade. The following equations summarizes the buy-sell conditions:
det
A y
B x
> 0 For the buy signal ... and (17)
det
A x
B y
>0For the sell signal ... ....... (18)
3.4 Production models
EVX indicator is accompanied by two production grade machine learning classi-
fication algorithms as complete python packages. The documentation on how to
apply the models is briefly described in the python package indexing page (PyPi).
The exact parameters used to obtain the draw-downs shown in this whitepaper and
achieve the desired backtesting metrics are considered proprietary, and thus are not
included in the free download pages.
4 case study
4.1 Assets screening
Volatility, industry exposure, market size, and trading liquidity among other com-
mon market risks forms the core of any well constructed portfolio. This is besides
the notion that the assets constituting the portfolio (stock universe) possesses a
highly non-linear correlation to each other, and thus to the overall impact on the
case study 9
trading [5]. The algorithm needs to start somewhere. This initial phase entails ran-
domly selecting assets, downloading historical data and running the screening to
assess long-term performance of the asset. In this phase, the algorithm returns the
list of assets when subjected to extreme requirements of zero maximum draw down
(MDD) with a specified set stop-loss. Here is an example of selected assets from the
bitcoin market:
(a) Initial coin list. (b) Summary metrics.
(c) Sell reasons statistics
Figure 2: Trades simulation for screening.
The above results were obtained through trades simulation with a stop loss of 7.4%
and a starting capital of 0.15 BTC for 225 days.
4.2 Optimized assets selection
The second phase involves knocking-off the assets that are susceptible to large draw
downs over a long period of time. Supposing that the the trader now decides to
remove the assets that are unproductive or by manipulating the stop loss range. He
will then remain with a list with zero-tolerance to losses as depicted in the figure
below:
(a) Intermediate coin list. (b) Summary metrics.
(c) Sell reasons statistics
Figure 3: Trades simulation for zero tolerance.
risk assessment and monitoring 10
5 risk assessment and monitoring
The most important part of any trading strategy is risk assessment. At DeFinance,
we have managed to leverage agile techniques to implement and monitor trading
risks and we continue to make improvements on a daily basis.
5.1 Draw down optimization
This phase involves setting allowable risk by the trader. Of importance here is
striking the balance between acceptable MDD and the instances of trades entered.
Setting a strictly low MDD means the trader does not enter into too many trades
or open too many positions at ago. While this minimizes the risk, it also reduces
the gains. However, with a certain acceptable range of MDD, the trader is able to
enter into as many trades as he may need. The following example shows trades
simulation under the same conditions, but the trader has accepted some losses in
order to increase the trade volumes. This ends up with an MDD of only 19.73%
which is well below the generally accepted value of 20%:
(a) Final coin list. (b) Summary metrics.
(c) Sell reasons statistics
Figure 4: Optimized trades with acceptable risk allowance. Initial amount: 0.0184 BTC, posi-
tions opened: 4, testing time period 458 days (Nov. 2019 - Feb. 2021)
5.2 Other important ratios
Besides maximum draw-down, we also use conventional as well as newly devel-
oped risk assessment parameters such as Sharpe, Sortino, Burke, Calmar, Sterling
and profit-to-draw-down ratios. In addition, we have compiled a number of in-
house risk management methodologies in relation to the developed algorithm. The
evx-coefficient for instance, is an aggregation of above ratios such that:
α=PN
i=0ri
CV (19)
Where alpha is the evx-coefficient, CV is the coefficient of variability based on the
daily profit percentages, while rare individual representations of the risk assess-
ment ratios earlier described. The second aggregate beta, is obtained as a linear
combination of the ratios, with beta coefficients assigned on the basis of the ratio’s
risk assessment and monitoring 11
relative significance as follows:
B=PN
i=0βiri
Rdσ2365 (20)
In the above system, σ2is annualized variance (of daily profits), and Rd, the mean
daily returns percent. Coefficients βare assigned through some optimization rou-
tine such as hyperopt. These methods are especially important when comparing
strategies and portfolios. Further optimization is often necessary, for instance to get
the right combinations of the last two parameters, that yields the desired results.
A continuous evaluation of these parameters in addition to live-testing during the
pilot-scale project has enabled us to move from dismally performing equity to a
much promising system. The following images present plots of equity curve before
and after implementation of the described methods.
Figure 5: Equity before optimization
Figure 6: Equity after optimization
the business model 12
Where Equity, Eis evaluated as a cumulative integral of daily average returns, Rd
for the entire backtesting period, tas follows:
E(t)=Zt
0
Rd(t)dt (21)
6 the business model
6.1 DeFinance Economics
Figure 7: DeFinance Economics Illustration
There are a couple of reasons why a trader would choose to use the trading sys-
tem. Conventional(manual) trading not only expose the trader to potential loses
due to inept operations, but also excessive emotional strain and stress from trend
following that may not be necessarily accurate. DeFinance trading is envisaged as
a service provider aiming to shield the trader from immense volatility in the new
DeFi arena. The ecosystem involves the client depositing trade-able amount of as-
set(s) in our predefined exchange and linking the credentials namely, the API-key
and API-secret used to authenticate to the respective exchange(s). The final step is
for the operator to initiate buy (and sell) orders sending. This is only necessary if
the exchange doesn’t support direct client authentication. The bot thus orchestrates
continuous trend following and submission of the orders on behalf of the trader
whenever market imperfections are detected on a 24/7basis.
6.2 Subscription Basis
The very nature of continuous operations implies that the headless operation must
be deployed and hosted onto a distributed computing environment – cloud service.
As this involves an overhead cost to the service provider, this cost is transferred
along with the trading fees, and applicable taxes to the client. The service provider
then charges a set minimum subscription fee for overhead charges, and a commis-
sion based on the expected increase in the stake currency at the end of the agreed
contractual period. In this way, the service not only helps the trader to maximise the
returns from bullish trading, but also to increase trading volumes while reducing
the risks by avoid unnecessary trading during high volatility seasons.
summary 13
7 summary
We have presented a justified case for a new financial markets indicator and how it
can be leveraged to maximize gains while at the same time minimizing risks in the
volatile market of decentralized finance. It is worth noting however that backtesting
results do not necessarily directly translate to the future outcomes due to the nature
of assets unpredictability. While the algorithm has proven promising during the
proof of concept (POC) test period with beta users, there is still a need to exercise
caution on its applicability on your own as we continue improving its accuracy. For
further information concerning our operations, consultations on implementation,
and partnering negotiations, contact DeFinance R
team.
references
[1] Antonio C Briza and Prospero C Naval Jr. Stock trading system based on the
multi-objective particle swarm optimization of technical indicators on end-of-
day market data. Applied Soft Computing,11(1):11911201,2011.
[2] Anton Korinek et al. Systemic risk: Amplification effects, externalities, and
policy responses. Technical report, 2009.
[3] Daron Acemoglu, Asuman Ozdaglar, and Alireza Tahbaz-Salehi. Systemic risk
and stability in financial networks. American Economic Review,105(2):564608,
2015.
[4] Lei Fang, Jiang Cheng, and Fang Su. Interconnectedness and systemic risk:
A comparative study based on systemically important regions. Pacific-Basin
Finance Journal,54:147158,2019.
[5] Una-May O’Reilly, Tina Yu, Rick Riolo, and Bill Worzel. Genetic programming
theory and practice II, volume 8. Springer Science & Business Media, 2006.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
The worst financial crises since the Great Depression has forced central bankers and policymakers across Europe and around the globe to take unprecedented policy measures to deal with systemic risk, i.e. the risk that the financial system ceases to perform its function of allocating capital to the most productive use because of financial difficulties among a significant number of financial institutions. This paper develops a parsimonious model of systemic risk in the form of amplification effects whereby adverse developments in financial markets and in the real economy mutually reinforce each other and lead to a feedback cycle of falling asset prices, deteriorating balance sheets and tightening financing conditions. The paper shows that the free market equilibrium in such an environment is generically inefficient because constrained market participants do not internalize that their actions entail amplification effects. Therefore they undervalue the social benefits of liquidity during crises and take on too much systemic risk. We use our framework to shed light on a number of current policy issues. We show that banks face socially insufficient incentives to raise more capital during systemic crises, that bailouts which are anticipated can be ineffective, and that expectational errors are considerably more costly during crises than in normal times. Furthermore we develop an analytical framework for macro-prudential capital adequacy requirements that take into account systemic risk. We also analyze a new channel of financial contagion and explain why private agents will take insufficient precautions against contagion from other sectors in the economy. JEL classification: E31, E32, E24, J51 ;
Article
Full-text available
Stock traders consider several factors or objectives in making decisions. Moreover, they differ in the importance they attach to each of these objectives. This requires a tool that can provide an optimal tradeoff among different objectives, a problem aptly solved by a multi-objective optimization (MOO) system.This paper aims to investigate the application of multi-objective optimization to end-of-day historical stock trading. We present a stock trading system that uses multi-objective particle swarm optimization (MOPSO) of financial technical indicators. Using end-of-day market data, the system optimizes the weights of several technical indicators over two objective functions, namely, percent profit and Sharpe ratio.The performance of the system was compared to the performance of the technical indicators, the performance of the market, and the performance of another stock trading system which was optimized with the NSGA-II algorithm, a genetic algorithm-based MOO method. The results show that the system performed well on both training and out-of-sample data. In terms of percent profit, the system outperformed most, if not all, of the indicators under study, and, in some instances, it even outperformed the market itself. In terms of Sharpe ratio, the system consistently performed significantly better than all the technical indicators. The proposed MOPSO system also performed far better than the system optimized by NSGA-II.The proposed system provided a diversity of solutions for the two objective functions and is found to be robust and fast. These results show the potential of the system as a tool for making stock trading decisions.
Article
We apply a novel technique to identify systemically important regions (SIRs) in a global network that shows a reduced degree of concentration and the development of a multi-centered structure. We observe that when a region is more connected to other regions, it is exposed to a higher level of systemic risk. This condition holds even more strongly for non-systemically important regions. However, for SIRs, interconnectedness is not significantly associated with systemic risk. Our empirical evidence suggests that an increase in interconnectedness at the regional level, together with a decrease in interconnectedness for a single pivotal center, may reduce the aggregate systemic risk at the global level.
Article
We provide a framework for studying the relationship between the financial network architecture and the likelihood of systemic failures due to contagion of counterparty risk. We show that financial contagion exhibits a form of phase transition as interbank connections increase: as long as the magnitude and the number of negative shocks affecting financial institutions are sufficiently small, more “complete” interbank claims enhance the stability of the system. However, beyond a certain point, such interconnections start to serve as a mechanism for propagation of shocks and lead to a more fragile financial system. We also show that, under natural contracting assumptions, financial networks that emerge in equilibrium may be socially inefficient due to the presence of a network externality: even though banks take the effects of their lending, risk-taking and failure on their immediate creditors into account, they do not internalize the consequences of their actions on the rest of the network.Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.