The document discusses a study analyzing high frequency trading data from 29 Russian stocks over 6 months in 2010. 369 shock events were identified, where the price changed significantly within a short period. On average there were 13 shock events identified per stock. The study uses high frequency tick data and various filters to identify periods of abrupt price changes at microsecond to minute time scales. The goal is to better understand dynamics and inefficiencies in short term price formation.

2. About Prognoz
Leading Russian developers of Business Intelligence
and Performance Management systems
• international company that has
been working in the IT market
since 1991
• joint team of over 1 200 skilled
economists, programmers,
analysts
• 50% market of BI in Russia
• Prognoz Platform, 1-st Russian
platform in Magic Quadrant of
Gartner

3. CONTENTS
Technical architecture Practical approach
 Evolution of bubble and risk management
 About MMP cluster  Monitoring of financial bubbles
 MMP cluster architecture  The system of bubble recognition
Financial bubbles Science and experiment
 Historical bubbles  Financial bubble experiment
 Definition of financial bubbles  Market microstructure approach
Theory of crashes
 LPPL model
 Fitting of the model
 Models selection
3

4. Financial engineering
(Stylized facts)
……………………………………….…................…
Liquidity of the financial market and
assets
……………………………………………………….…..
Agent-based modeling and simulation
……………………………………………………………
Market microstructure analysis
…………………………..………………………………
Bubble detection and diagnosis
……………………………………………………………
4

5. Technical info:
 Installation Site: Perm state university
 Supercomputer type: Cluster
 Number of nodes: 3
 Number of Cores per node: 12
 CPU type: Intel Xeon 5650 (2.66 GHz)
 RAM per node: 64 Gb
 OS: Windows Server 2003
5

6. 6
Total: 48 services, 72 CPU, 228 Gb RAM

7.  R is statistical and graphical
programming environment
 Appeared in 1993 and designed by There is more than 4300
Ross Ihaka and Robert Gentleman packages that allow to use
specialized statistical
 R is a GNU project techniques, graphical
devices, import/export
 R – a free implementation of the S capabilities, reporting tools,
etc.
language
 It runs on a variety of platforms
including Windows, Unix and MacOS
 It contains advanced statistical
routines not yet available in other
packages
7

8. Commands
Database
Batch file R file
Task R file
Batch file
Runner
Batch file R file
8

9. 9

10. 10

11. 11

13. Mr. Greenspan
Thefreedictionary.com
Charles Kindleberger, MIT
Professor J.Barley Rosser, James Madison University
13

14. Authors
A.Johansen, O.Ledoit, D.Sornette (JLS)
First publication
Large financial crashes (1997)
Famous book
Didier Sornette
Why Stock Markets Crash (2004)
𝑡 𝑐 - critical time when bubble crash or
change to another regime
14

15. 𝑚
𝑙𝑛 𝑝 𝑡 = 𝐴 + 𝐵(𝑡 𝑐 − 𝑡)
𝑡𝑐
15

16. 𝐶(𝑡 𝑐 − 𝑡) 𝑚 𝑐𝑜𝑠[𝜔 𝑙𝑜𝑔 𝑡 𝑐 − 𝑡 − 𝜑]
16

17. 𝑙𝑛 𝑝 𝑡 = 𝐴 + 𝐵(𝑡 𝑐 − 𝑡) 𝑚 +𝐶(𝑡 𝑐 − 𝑡) 𝑚
𝑐𝑜𝑠[𝜔 𝑙𝑜𝑔 𝑡 𝑐 − 𝑡 − 𝜑]
17

18. 𝑙𝑛[𝑝(𝑡)] 𝑙𝑛 𝑝 𝑡 = 𝐴 + 𝐵(𝑡 𝑐 − 𝑡) 𝑚 +𝐶(𝑡 𝑐 − 𝑡) 𝑚 𝑐𝑜𝑠[𝜔 𝑙𝑜𝑔 𝑡 𝑐 − 𝑡 − 𝜑]
m = 0.01 m = 0.3
m = 0.9 m = 1.7
18

19. 𝑙𝑛 𝑝 𝑡 = 𝐴 + 𝐵(𝑡 𝑐 − 𝑡) 𝑚 +𝐶(𝑡 𝑐 − 𝑡) 𝑚
𝑐𝑜𝑠[𝜔 𝑙𝑜𝑔 𝑡 𝑐 − 𝑡 − 𝜑]
=3 =7
𝑡𝑐 − 𝑡
𝑡𝑐 − 𝑡 𝑡𝑐 − 𝑡
 = 15  = 30
𝑡𝑐 − 𝑡 𝑡𝑐 − 𝑡 19

20. 𝑙𝑛 𝑝 𝑡 = 𝐴 + 𝐵(𝑡 𝑐 − 𝑡) 𝑚 +𝐶(𝑡 𝑐 − 𝑡) 𝑚
𝑐𝑜𝑠[𝜔 𝑙𝑜𝑔 𝑡 𝑐 − 𝑡 − 𝜑]
=7  = 9.5
20

21. For each log periodic curve we fixed:
𝑡0 - start time of the bubble
First model 𝑡 𝑐 - critical time when bubble crash or
change to another regime
Second model
Sample of 𝑡 𝑐
𝑡 𝑐1 𝑡 𝑐2
21

22. John von Neumann
22

23. • Main filtration (0<m<1, B<0)
• Residuals stationarity tests (ADF test, Phillips–Perron test)
• Lomb spectral analysis
LOMB PERIODOGRAM
150
m
100
P(omega)
50
0
0 10 20 30 40
omega
23

24. Sample of
𝑡𝑐
Distribution
of 𝑡 𝑐
Quantiles
Risk measure
24

25. 25
25

26. D.Fantazzini, P.Geraskin,
Everything You Always Wanted to Know
about Log Periodic Power Laws for Bubble Modelling
but Were Afraid to Ask (2011)
26

27. Timeframe LPPL
• Bubble • Long • Large
• Anti - bubble • Short • Small • Parameters
Type Size
27

28. The Financial Crisis Observatory (FCO) is a scientific platform aimed at testing and
quantifying rigorously, in a systematic way and on a large scale the hypothesis that
financial markets exhibit a degree of inefficiency and a potential for predictability,
especially during regimes when bubbles develop. (http://www.er.ethz.ch/fco/index)
Testing two hypotheses:
• Hypothesis H1: financial (and other) bubbles can be diagnosed in real-time
before they end..
• Hypothesis H2: The termination of financial (and other) bubbles can be
bracketed using probabilistic forecasts, with a reliability better than chance
(which remains to be quantied).
D. Sornette, R. Woodard,
M. Fedorovsky,S. Reimann, H. Woodard, W.-X. Zhou
The Financial Bubble Experiment. First Results (2 November 2009 - 1 May 2010)
28

29.  2 November 2009 – 1 May 2010 [http://www.er.ethz.ch/fco/FBE_report_May_2010]
 2 of 4 bubbles detected by model were real bubbles
 All of them changed their regimes
 12 May 2010 – 1 November 2010 [http://www.er.ethz.ch/fco/fbe_Report_1Nov10_2]
 5 of 7 bubbles detected by model were real bubbles
 4 of 5 changed their regimes
 12 November 2011 – 2 May 2011 [http://www.er.ethz.ch/fco/fbe_20110502_assets_3.pdf]
 24 of 27 bubbles detected by model were real bubbles
 17 of 24 changed there regime
29

30. NBER Working Group
30

31. Different types of filters at 3 time scales:
 Hours scale (macro):
 Absolute filter
 Relative filter
Source: Guo-Hua Mu, Wei-Xing Zhou, Wei Chen and J´anos Kert´esz. Order flow dynamics around extreme
price changes on an emerging stock market, 2010
 Minutes scale (meso):
 Filter of minute returns
Source: Armand Joulin, Augustin Lefevre, Daniel Grunberg, Jean-Philippe Bouchaud. Stock price jumps:
news and volume play a minor role, 2010
 Tick scale (micro):
 NANEX filter
Source: Flash Crash Analysis Continuing Developments
http://www.nanex.net/FlashCrashEquities/FlashCrashAnalysis_Equities.html
62
1.855
1.845 61.5
1.835
61
1.825
60.5
price [rub]
price, rub.
1.815
60
1.805
59.5
1.795
1.785 59
1.775 58.5
31
12:06:00
12:18:00
12:30:00
12:42:00
12:54:00
14:06:00
14:18:00
14:30:00
14:42:00
14:54:00
16:06:00
16:18:00
16:30:00
16:42:00
16:54:00
17:54:00
18:06:00
18:18:00
18:30:00
11:30:00
11:42:00
11:54:00
13:06:00
13:18:00
13:30:00
13:42:00
13:54:00
15:06:00
15:18:00
15:30:00
15:42:00
15:54:00
17:06:00
17:18:00
17:30:00
17:42:00
11:32:20
11:59:04
12:41:35
12:56:35
13:07:30
13:39:18
14:15:14
14:38:45
14:57:21
15:28:13
16:04:43
16:28:36
17:35:22
18:31:01
11:42:00
11:52:45
12:07:45
12:24:08
12:50:44
12:53:52
13:01:54
13:04:06
13:24:35
13:33:16
13:52:46
14:02:26
14:24:38
14:26:15
14:43:58
14:51:17
15:01:22
15:15:36
15:57:19
16:00:26
16:16:12
16:23:20
16:35:55
17:03:44
17:46:30
18:02:16
time
time [ticks]

32. for 16 ticks
(61.2 ->61.7)
Within 1 second
price rose 0.82 %
price [rub]
61.5
58.5
59.5
60.5
59
60
61
62
11:32:20
11:42:00
11:52:45
11:59:04
12:07:45
12:24:08
12:41:35
12:50:44
12:53:52
12:56:35
13:01:54
13:04:06
13:07:30
13:24:35
13:33:16
13:39:18
13:52:46
14:02:26
14:15:14
14:24:38
14:26:15
14:38:45
time [ticks]
14:43:58
14:51:17
14:57:21
15:01:22
15:15:36
15:28:13
15:57:19
16:00:26
16:04:43
16:16:12
16:23:20
16:28:36
16:35:55
17:03:44
17:35:22
32
17:46:30
18:02:16
18:31:01

33. Statistics: IDENT UP DOWN ALL
PMTL 15 36 51
MAGN 31 6 37
Stocks analyzed 29 blue chips NOTK
OGKC
18
13
18
23
36
36
01.04.2010-30.06.2010; AFLT 9 25 34
RTKM 14 19 33
Period 1.09.2010-12.10.2010 MGNT 4 16 20
NLMK 8 12 20
Trading days 82 URKA 7 11 18
Sample analyzed 20.2 mln. ticks SIBN
RASP
6
7
10
8
16
15
Trading time 11.30-18.40 MRKH 3 9 12
MSNG 5 7 12
CHMF 3 4 7
Shocks found 369 RU14TATN3006
HYDR
3
3
3
2
6
5
TRNFP 3 0 3
IUES 0 2 2
We use a tick dynamics of prices for MTSI 1 1 2
SNGSP 2 0 2
filtering (source: MICEX) ROSN 1 0 1
SNGS 1 0 1
FEES 0 0 0
GAZP 0 0 0
GMKN 0 0 0
Total 369 events (13 per stock) LKOH 0 0 0
On average 1 shock/7 days per stock SBER03 0 0 0
SBERP03 0 0 0
VTBR 0 0 0
Average 5 7 33
13

34. Science
Laboratory of financial modeling and risk
management - Prognoz Risk Lab
Мagistracy in finance and IT (Master in
Finance & IT) in Perm State National Research
University
mifit.ru
Perm Winter School is an annual conference
on modeling of financial markets and risk
management
permwinterschool.ru