An early indicator for anomalous stock market performance

Key Question Being Explored in This Insight:

How can a real-time indicator detect over 80% of stock market anomalies, including the 1987 Bubble and 2007 Financial Crisis?

View the page of: Thomas Gries

Institution: Universität Paderborn

Explore more insights in:

Economics Financial Economics Social Sciences

Media Literacy Toolkit

Faculti's Media Literacy Toolkit helps viewers critically engage with academic insights by analyzing the research context, identifying perspectives, and encouraging thoughtful evaluation.

Critical Questions to Consider ▼

What assumptions does the research make?
Are there alternative perspectives not explored?
What are the limitations of the research method?

Bias and Perspective Awareness ▼

This research comes from Universität Paderborn . Reflect on how the institution's academic focus and research partnerships may shape the questions being explored.

Further Reading ▼

Read the Study

How can a real-time indicator detect over 80% of stock market anomalies, including the 1987 Bubble and 2007 Financial Crisis?

Image courtesy of interviewee. December 17, 2024

Log-in or Sign-up to Faculti
Currently viewing this subject insight as a guest. You have insight(s) remaining for this month. Login to view 8000+ figures on the platform.

Generating Workspace...

Export as PDF

This video video discusses the challenge of identifying and predicting financial market anomalies, such as bubbles and crashes. The authors highlight the lack of consensus in existing theories regarding what constitutes a bubble, how it develops, and whether it can be predicted. To tackle this issue, they propose a data-driven approach to analyze market behaviors through time series analysis, specifically examining the S&P 500 and utilizing earnings-to-price ratios as indicators of market conditions.

The methodology involves decomposing time series data to establish what is "normal" market behavior, using techniques like local linear trend estimation instead of arbitrary parameters. The authors introduce a statistical framework characterized by the detection of trends and the analysis of residuals using Autoregressive Conditional Heteroskedasticity (ARCH) models, allowing for the identification of systematic movements in the data.

The results indicate the ability to predict when the market might transition from normal to anomalous conditions, with an 80% detection rate of identified anomalies from 1970 to 2020. The authors emphasize that while they cannot predict specific market crashes, they can signal potential shifts in market behavior that warrant caution, enabling market participants to adjust their strategies accordingly. The potential for refining this predictive methodology through theoretical integration and improved modeling is acknowledged, highlighting the need for further research into market dynamics and the associated risks of anomalous episodes.

Language:

Export as PDF

Export as PDF
Here are five key points from the video:

1. **Challenge of Identifying Financial Bubbles**: The video discusses the difficulty in identifying and predicting financial bubbles and crashes, as there is no consensus on what constitutes a bubble and how to recognize it prior to its emergence.

2. **Focus on Statistical Analysis**: The authors propose a purely statistical approach to analyze market behaviors. They aim to determine what constitutes "normal" and "abnormal" market conditions using time series analysis, specifically looking at earnings to price ratios as indicators.

3. **Methodological Innovations**: The analysis employs a novel method of trend detection using an endogenous local linear approach, which is data-driven and avoids the need for arbitrary parameters. This is seen as an improvement over traditional methods like the HP filter.

4. **Detection of Anomalous Conditions**: The methodology allows for the identification of systematic movements in market data, particularly through the use of autoregressive models (CTAs), to predict potential market anomalies. The results indicate an ability to detect 80% of existing anomalies, which could signal the onset of potential bubbles.

5. **Warning Signals and Market Behavior**: The findings suggest that by identifying anomalous market conditions, agents can receive early warnings about potential investment risks, enabling them to make more informed decisions about their market exposure. The authors advocate for market participants to adopt caution instead of relying on external regulatory interventions.

Export as PDF
Here are some key quotes extracted from the video you :

1. **Difficulty in Identifying Bubbles:**
- "We had the question why nobody really can identify or even predict what we sometimes perceive as a bubble or pressure in financial markets."

2. **Lack of Consensus in Bubble Theory:**
- "There’s no really consensus in theory what the bubble actually is and what the crash is."

3. **Normal vs. Abnormal Market Behavior:**
- "The question we had is, is it possible to identify normal market behavior and something like a normal market behavior?"

4. **Statistical Analysis Approach:**
- "What you do in time series analysis is that you start to decompose your time series."

5. **Using Earning to Price Ratios:**
- "The time series we looked at here was the earning to price ratios (AP ratios) as one of the standard indicators that may give us some idea about the market conditions."

6. **Identification of Trends:**
- "We introduce an endogenous local linear trend... it’s fully data-driven and doesn’t need some kind of arbitrary parameters."

7. **Identifying Systematic Movements:**
- "After detecting the trend, we looked at the residuals and then tried to find out is there something systematic in the residuals."

8. **Understanding Predicted Anomalies:**
- "We did not want to predict… what we can predict maybe is that the market is on the way to turn anomalous…"

9. **Warning Signals for Market Agents:**
- "Maybe there’s a crash ahead… Therefore, market agents should become a little bit more careful in their decisions."

10. **Historical Anomalies Identified:**
- "The good thing is that with this method, we could identify these kind of anomalous episodes that at the end turned out to be bubbles like the bubble and the crash in 1987."

11. **Possible Interventions:**
- "We do not think it makes a big sense that some kind of agency should step in... however, if the market realizes something is unusual..."

12. **Future Improvements:**
- "There are ways even to improve the trending method, but also... combine our purely statistical method here with some theory."

These quotes highlight the issues regarding predicting bubbles, the methodologies used, and the implications for market participants.

X-TIMESTAMP-MAP=LOCAL:00:00:00.000,MPEGTS:90000

1
00:00:00.105 --> 00:00:03.205
We had the question why nobody really can identify

2
00:00:03.305 --> 00:00:06.925
or even predict, what we sometimes perceive as a,

3
00:00:06.925 --> 00:00:09.725
as a bubble or pressure in financial markets.

4
00:00:10.465 --> 00:00:12.085
Uh, so therefore we,

5
00:00:12.105 --> 00:00:14.365
we couldn't really understand why, why that's so difficult.

6
00:00:14.425 --> 00:00:16.805
And therefore, we started to think about that

7
00:00:17.925 --> 00:00:21.745
and try to find out if there we can anything,

8
00:00:21.845 --> 00:00:25.985
do in order to, get an idea what happens there.

9
00:00:26.085 --> 00:00:30.905
And, maybe, that it is even early to detect.

10
00:00:31.045 --> 00:00:34.105
We started to look actually at, at the bubble literature

11
00:00:34.105 --> 00:00:35.985
and, and knew about the bubble literature,

12
00:00:35.985 --> 00:00:39.145
but there's no really consensus in, in theory

13
00:00:39.295 --> 00:00:41.665
what the bubble actually is and what the crash is.

14
00:00:41.725 --> 00:00:45.545
And, so the only thing which people find is,

15
00:00:45.605 --> 00:00:49.105
and, that's also what we, what we see in public,

16
00:00:49.105 --> 00:00:50.985
that at the end with the crash, there's a,

17
00:00:50.985 --> 00:00:53.585
there's a huge damage, but what the bubble is

18
00:00:53.585 --> 00:00:55.345
and how we get into the bubble

19
00:00:55.485 --> 00:00:58.585
and, if it is predictable even, is something which,

20
00:00:58.715 --> 00:01:00.545
where there's no really theory

21
00:01:00.725 --> 00:01:04.105
and no really, clear consensus about that.

22
00:01:04.525 --> 00:01:08.525
So therefore, our idea was that we may have

23
00:01:08.525 --> 00:01:12.325
to look at purely at data and then try to analyze the data,

24
00:01:12.425 --> 00:01:14.605
and maybe we find out something about the data.

25
00:01:15.265 --> 00:01:18.765
And therefore, the question we had is, is it possible

26
00:01:18.785 --> 00:01:20.885
to identify normal market behavior

27
00:01:21.185 --> 00:01:23.285
and something like a normal market behavior?

28
00:01:24.105 --> 00:01:26.685
So normal or abnormal was the question here.

29
00:01:27.265 --> 00:01:30.245
Um, or is it possible even to predict early enough

30
00:01:30.675 --> 00:01:34.245
that resist some, some kind of future period, maybe a year?

31
00:01:34.865 --> 00:01:38.485
Um, the market will cross from a normal to a,

32
00:01:38.665 --> 00:01:40.285
an anomalous condition,

33
00:01:40.745 --> 00:01:44.565
and that is something we may identify purely statistically.

34
00:01:45.845 --> 00:01:47.425
So then the question is, what is normal?

35
00:01:47.455 --> 00:01:48.585
What is un-normal, of course,

36
00:01:48.605 --> 00:01:50.905
or what is a normal unnormal behavior?

37
00:01:51.685 --> 00:01:54.665
And again, we go to pure statistics.

38
00:01:54.725 --> 00:01:58.265
And what you do in, time series analysis is

39
00:01:58.265 --> 00:02:02.505
that you start to decompose your, your time series.

40
00:02:02.645 --> 00:02:05.065
And the time series we looked at here was the

41
00:02:05.625 --> 00:02:06.745
standard and poor 500.

42
00:02:07.485 --> 00:02:10.865
And, here we looked at the earning to price ratios,

43
00:02:11.045 --> 00:02:14.345
AP ratios as, one of the standard indicators

44
00:02:14.415 --> 00:02:16.985
that may give us some idea about,

45
00:02:17.605 --> 00:02:18.745
the market conditions.

46
00:02:18.925 --> 00:02:23.025
Uh, it was very simple actually, what we could define

47
00:02:23.045 --> 00:02:24.345
as normal or un-normal.

48
00:02:24.845 --> 00:02:26.145
Um, so as known

49
00:02:26.325 --> 00:02:30.145
and even well known from something like the Bollinger Bands,

50
00:02:30.725 --> 00:02:33.105
we identify the trend movements

51
00:02:34.085 --> 00:02:37.745
and then defined a band around these trend movements

52
00:02:37.745 --> 00:02:41.305
of about 95% of all observations,

53
00:02:41.495 --> 00:02:42.945
that might be in this band.

54
00:02:43.605 --> 00:02:46.345
And then the economic interpretation is

55
00:02:46.415 --> 00:02:49.545
that the trend is something like the fundamental market

56
00:02:49.945 --> 00:02:53.185
movement and, that there is some, some kind of other

57
00:02:53.755 --> 00:02:57.105
systematic or random movement around, we identified that

58
00:02:57.125 --> 00:02:58.585
as long the market is within

59
00:02:58.655 --> 00:03:03.005
This band, we see that we are under kind

60
00:03:03.005 --> 00:03:06.125
of normal behavior of the market under normal conditions.

61
00:03:06.865 --> 00:03:10.685
Um, however, if the market, is moving out of the bands,

62
00:03:10.835 --> 00:03:12.525
then it turns to be anomalous.

63
00:03:13.145 --> 00:03:16.685
And that is something we can purely look, up from the

64
00:03:17.285 --> 00:03:19.605
statistics and by statistic definitions.

65
00:03:20.225 --> 00:03:22.045
So up to that point, nothing is really new.

66
00:03:22.735 --> 00:03:24.665
However, the novelty here

67
00:03:24.665 --> 00:03:26.545
and now comes when we look at the methods,

68
00:03:26.725 --> 00:03:28.065
how we do the decomposition,

69
00:03:29.165 --> 00:03:32.545
and, let's start maybe with a, with a, with a trend.

70
00:03:33.525 --> 00:03:37.025
So we look at the trend and for the trend,

71
00:03:37.045 --> 00:03:39.665
and the identification of the trend,

72
00:03:39.965 --> 00:03:42.705
we introduce an endogenous local linear

73
00:03:42.835 --> 00:03:44.185
trend estimation approach.

74
00:03:44.455 --> 00:03:48.225
This is innovative, because it's fully data driven

75
00:03:48.365 --> 00:03:50.985
and doesn't need, some, some kind

76
00:03:50.985 --> 00:03:53.385
of arbitrary parameters, which we have to put in.

77
00:03:54.045 --> 00:03:55.145
And, these kind

78
00:03:55.145 --> 00:03:57.465
of methodology actually was developed earlier

79
00:03:58.205 --> 00:04:00.065
in several papers, in a group

80
00:04:00.065 --> 00:04:03.905
of researchers at this university, namely Huang

81
00:04:03.905 --> 00:04:05.505
and Marlon Fit, and myself

82
00:04:05.505 --> 00:04:08.425
and some others were involved in this, development of,

83
00:04:08.425 --> 00:04:10.025
of, of this methodology of this.

84
00:04:10.525 --> 00:04:12.865
And ous,

85
00:04:13.275 --> 00:04:15.665
trend methodology is extremely efficient.

86
00:04:16.405 --> 00:04:20.905
Um, and, the trend can be identified easily,

87
00:04:21.805 --> 00:04:25.825
better than, some, some other trend methods,

88
00:04:25.825 --> 00:04:29.825
because we, again, use statistical criteria in order to,

89
00:04:29.825 --> 00:04:32.425
get to the, the optimal,

90
00:04:33.825 --> 00:04:35.185
determination of, of this trend.

91
00:04:35.685 --> 00:04:40.265
So it is, a method which is better than the often used

92
00:04:40.805 --> 00:04:42.665
HP filter method or some other method.

93
00:04:42.735 --> 00:04:44.705
Therefore, the, this new,

94
00:04:44.705 --> 00:04:48.065
innovative trend method is one important ingredient,

95
00:04:48.525 --> 00:04:49.585
of our analysis.

96
00:04:50.405 --> 00:04:55.105
Um, so the second, methodology we used,

97
00:04:55.685 --> 00:04:59.025
after detecting the trend,

98
00:04:59.025 --> 00:05:00.265
we looked at the residuals

99
00:05:00.285 --> 00:05:01.305
and then tried

100
00:05:01.305 --> 00:05:04.105
to find out is there something systematic in the residuals?

101
00:05:04.725 --> 00:05:07.865
And, actually we did that with Aceta model

102
00:05:08.565 --> 00:05:13.555
and these kind of C models, really could identify,

103
00:05:13.555 --> 00:05:16.555
systematic movements apart from the trend.

104
00:05:17.295 --> 00:05:21.675
And that helped, of course, in the third step, to

105
00:05:22.365 --> 00:05:25.555
model good prediction models in which we,

106
00:05:25.735 --> 00:05:30.155
now in the third step could do predictions, both

107
00:05:30.615 --> 00:05:33.555
for the normal regions

108
00:05:33.575 --> 00:05:35.915
and the development of the normal region of the market,

109
00:05:36.495 --> 00:05:38.355
but also for the abnormalities

110
00:05:38.695 --> 00:05:41.475
and for also the, the market itself.

111
00:05:41.855 --> 00:05:44.075
So with the help of, the CER

112
00:05:44.075 --> 00:05:48.075
and these, local linear trend methodology,

113
00:05:48.175 --> 00:05:51.075
and both using in the prediction models,

114
00:05:51.095 --> 00:05:53.075
we had some novelties in our methodology

115
00:05:53.775 --> 00:05:56.875
and that, probably helped to get the good

116
00:05:56.925 --> 00:05:58.685
Performance of the indicator at the end.

117
00:05:59.025 --> 00:06:00.405
CTA models is quite important.

118
00:06:00.705 --> 00:06:02.565
CTA model is an auto aggressive model.

119
00:06:03.225 --> 00:06:05.805
Um, so what we do is we identify,

120
00:06:05.835 --> 00:06:07.765
that there is in the residual.

121
00:06:07.985 --> 00:06:10.845
So after the det trending method,

122
00:06:10.845 --> 00:06:11.925
we look at the residuals

123
00:06:11.925 --> 00:06:14.325
and the residuals actually,

124
00:06:14.395 --> 00:06:16.965
they have some systematic movements

125
00:06:17.025 --> 00:06:19.325
and these systematic systematic movements,

126
00:06:19.915 --> 00:06:23.125
they can be picked up by the, CA model.

127
00:06:23.865 --> 00:06:27.705
And, therefore with the CT a model, we

128
00:06:28.295 --> 00:06:32.905
also get the improvement that the, direction,

129
00:06:33.645 --> 00:06:35.625
is, can be asymmetric.

130
00:06:35.805 --> 00:06:38.945
So the movements upwards may be different than

131
00:06:38.945 --> 00:06:40.025
the movements downwards.

132
00:06:40.845 --> 00:06:43.985
Um, and, for each direction, for the upwards

133
00:06:43.985 --> 00:06:46.385
and downward movements, we get a,

134
00:06:46.765 --> 00:06:48.985
an extra kind of modeling.

135
00:06:49.805 --> 00:06:51.545
Um, and that can be used also

136
00:06:51.545 --> 00:06:52.945
for the predictions at the end.

137
00:06:52.965 --> 00:06:56.105
So therefore, these, the CA model,

138
00:06:56.175 --> 00:06:57.745
goes in both directions

139
00:06:57.765 --> 00:07:00.465
and identified the systematic market movements in both

140
00:07:00.465 --> 00:07:03.385
directions, even if they are sym asymmetric.

141
00:07:03.455 --> 00:07:06.985
This asymetry is something which we also identified,

142
00:07:07.005 --> 00:07:09.505
and we identified the systematic movement.

143
00:07:10.085 --> 00:07:12.905
And this systematic movement could be used later on,

144
00:07:13.445 --> 00:07:14.665
for the predictions.

145
00:07:14.885 --> 00:07:18.585
Um, and the indicator, what we did not want

146
00:07:18.585 --> 00:07:21.945
to do is we did not want to predict that,

147
00:07:22.045 --> 00:07:23.545
in three months time, there's a crash.

148
00:07:24.245 --> 00:07:27.665
Um, so that is nothing, which is predictable actually.

149
00:07:28.525 --> 00:07:31.865
Um, however, what we can predict maybe is

150
00:07:31.865 --> 00:07:34.385
that the market is on the way to turn animals

151
00:07:35.285 --> 00:07:38.585
and, this may happen within the next year.

152
00:07:39.285 --> 00:07:42.345
And then once the market moves in an animal region,

153
00:07:42.975 --> 00:07:44.665
then the chance for turning point

154
00:07:44.765 --> 00:07:49.745
or market correction is clearly increasing.

155
00:07:50.645 --> 00:07:54.065
And so therefore, also it, what is increasing is

156
00:07:54.455 --> 00:07:56.745
that from these anomalous conditions,

157
00:07:57.125 --> 00:08:00.345
the market may even turn into something like a, a bubble

158
00:08:00.405 --> 00:08:02.865
or crash at the end, there may be a correction

159
00:08:03.445 --> 00:08:04.905
and a turning point ahead, and

160
00:08:04.905 --> 00:08:09.185
therefore, the first step is to identify when,

161
00:08:09.185 --> 00:08:10.545
market turns anomalous.

162
00:08:11.325 --> 00:08:13.625
And, this is what, what we try to do.

163
00:08:14.485 --> 00:08:17.545
And again, also with a, with a specification,

164
00:08:17.575 --> 00:08:20.825
with a choice of the specification of the CTA model,

165
00:08:21.045 --> 00:08:23.185
we use purely statistical criteria.

166
00:08:24.045 --> 00:08:27.185
And, with the help of these criteria,

167
00:08:27.445 --> 00:08:29.545
we were able then to detect

168
00:08:30.205 --> 00:08:33.265
if there are anomalies ahead.

169
00:08:34.085 --> 00:08:37.225
Um, and the, the rate, how we

170
00:08:38.025 --> 00:08:41.625
identified the quality, the capability of, detecting

171
00:08:41.625 --> 00:08:45.585
that is by the two positive rate and the two positive rate.

172
00:08:46.245 --> 00:08:49.105
Um, here was 80, 80%.

173
00:08:49.125 --> 00:08:54.065
So 80% of all the exposed with all information

174
00:08:54.785 --> 00:08:58.925
Existing anomalies, were detected, by,

175
00:08:58.985 --> 00:09:00.325
by our criteria.

176
00:09:00.625 --> 00:09:03.485
The question is, how and what could we detect there?

177
00:09:04.225 --> 00:09:09.205
Um, however, first, what, what it, what it means, if we,

178
00:09:09.365 --> 00:09:13.085
I identify something like 23, anomalies episodes,

179
00:09:13.905 --> 00:09:15.605
it means that from time to time,

180
00:09:15.665 --> 00:09:17.765
the market moves into something unusual.

181
00:09:18.065 --> 00:09:20.845
So it's a simply statistical idea.

182
00:09:21.025 --> 00:09:22.725
It is something which is unusual.

183
00:09:22.725 --> 00:09:25.925
However, this may not be a problem.

184
00:09:26.145 --> 00:09:27.805
So the market may correct somehow

185
00:09:28.625 --> 00:09:32.805
and turn back to, to normal after this unusual episode.

186
00:09:33.465 --> 00:09:36.085
And most of the time, this happens actually this way.

187
00:09:36.745 --> 00:09:41.605
Um, however, there may be also something,

188
00:09:41.785 --> 00:09:43.645
as a result of this episode,

189
00:09:43.645 --> 00:09:45.285
which at the end is harmful.

190
00:09:45.545 --> 00:09:47.885
So normally it's, not really harmful,

191
00:09:48.105 --> 00:09:52.805
but, in some cases, these anom,

192
00:09:52.945 --> 00:09:55.685
anomalous conditions, turn out to

193
00:09:56.585 --> 00:10:01.205
be something which we later perceive as bubbles and crashes.

194
00:10:01.945 --> 00:10:06.125
So therefore, detecting anomalous episodes is,

195
00:10:06.235 --> 00:10:08.285
something which in the first moment is,

196
00:10:08.285 --> 00:10:09.965
is not really a big thing.

197
00:10:10.585 --> 00:10:14.825
Um, however, maybe it turns into a bubble in a crash,

198
00:10:15.325 --> 00:10:16.465
and therefore it makes sense

199
00:10:16.525 --> 00:10:20.065
to give a signal in the market a warning signal in the

200
00:10:20.065 --> 00:10:23.145
market here, something which is anomalous and

201
00:10:23.145 --> 00:10:25.465
therefore maybe there's a crash ahead.

202
00:10:25.465 --> 00:10:29.545
Therefore, you view the market, the market agents,

203
00:10:30.245 --> 00:10:32.505
should, become

204
00:10:33.145 --> 00:10:35.625
a little bit more careful in your decisions and, and,

205
00:10:35.685 --> 00:10:38.465
and check if you really wanna stay in that market.

206
00:10:39.325 --> 00:10:40.785
And, so

207
00:10:40.785 --> 00:10:45.465
therefore, what we discovered here is, 70

208
00:10:45.805 --> 00:10:48.905
or 80% of these 23 anomalous conditions.

209
00:10:49.725 --> 00:10:53.825
Um, it started in 1970 to 2020,

210
00:10:54.925 --> 00:10:59.745
in 1970 because, we needed the time from,

211
00:10:59.925 --> 00:11:01.745
before 1970,

212
00:11:01.745 --> 00:11:04.945
after the second World War, in order to adjust the model,

213
00:11:05.125 --> 00:11:08.545
the model needs also some, some time before,

214
00:11:08.545 --> 00:11:11.505
however, afterwards, with the two positive rate,

215
00:11:11.505 --> 00:11:12.745
80% were detected.

216
00:11:13.085 --> 00:11:17.065
And, now the idea is that, the

217
00:11:18.415 --> 00:11:21.305
indicator may, may signal in the market, yes,

218
00:11:21.305 --> 00:11:23.105
there's something unusual

219
00:11:23.765 --> 00:11:26.025
and maybe even something like a bubble ahead.

220
00:11:26.405 --> 00:11:29.545
And the good thing is that with this method, we could

221
00:11:30.145 --> 00:11:32.545
identify these kind of anomalous episodes

222
00:11:33.095 --> 00:11:36.865
that at the end turned out to be bubbles like the bubble

223
00:11:36.885 --> 00:11:38.345
and the crash in 1987.

224
00:11:39.365 --> 00:11:43.345
Uh, we could, identify this, one year ahead

225
00:11:43.655 --> 00:11:47.705
with our predictions, and the same also for the 2008 bubble

226
00:11:47.765 --> 00:11:49.745
and crash and overvaluation.

227
00:11:50.405 --> 00:11:53.685
Uh, so if these, indicator

228
00:11:54.915 --> 00:11:58.885
were usable before, at least a year ahead

229
00:11:58.905 --> 00:12:02.245
or something, we could have given signals in the market

230
00:12:02.425 --> 00:12:05.565
so that, something unusual is going on.

231
00:12:06.065 --> 00:12:09.485
And with this, something unusual going on,

232
00:12:09.705 --> 00:12:14.045
the market could have decided to turn back to normal.

233
00:12:14.385 --> 00:12:18.405
And we may have not seen, the, the,

234
00:12:18.545 --> 00:12:19.845
the crashes and the bubbles.

235
00:12:19.945 --> 00:12:21.045
Um,

236
00:12:21.045 --> 00:12:23.285
however that, that is the idea,

237
00:12:23.925 --> 00:12:27.045
a warning signal in order to give warnings.

238
00:12:27.105 --> 00:12:29.445
And, and, there's something like an

239
00:12:30.265 --> 00:12:32.765
market correction from the market itself.

240
00:12:32.865 --> 00:12:36.845
We do not, think it makes a big sense that,

241
00:12:36.955 --> 00:12:39.165
some kind of agency should step in

242
00:12:39.165 --> 00:12:40.965
and do something if the signal comes up.

243
00:12:41.545 --> 00:12:45.765
Um, however, if if the market realizes something,

244
00:12:46.355 --> 00:12:50.005
unusual is ahead, then the market can decide itself.

245
00:12:50.145 --> 00:12:52.285
Um, is it unusual

246
00:12:52.285 --> 00:12:56.125
because, it should be unusual at this point,

247
00:12:56.745 --> 00:12:57.765
or is it unusual?

248
00:12:58.065 --> 00:13:02.245
And the market should, be more carefully think about,

249
00:13:02.425 --> 00:13:06.285
what, what decisions they make, if they wanna stay and,

250
00:13:06.285 --> 00:13:09.005
and right maybe on the, on the market movements,

251
00:13:09.585 --> 00:13:12.645
or should they really, be more careful and,

252
00:13:12.705 --> 00:13:14.845
and, check

253
00:13:15.425 --> 00:13:17.605
if they are still on the, on the right side.

254
00:13:18.185 --> 00:13:21.965
Uh, so therefore the, the signal should help the market,

255
00:13:21.965 --> 00:13:22.965
to give an orientation.

256
00:13:23.865 --> 00:13:26.965
Um, but there's no really particular

257
00:13:26.985 --> 00:13:29.085
policy involved in that.

258
00:13:29.265 --> 00:13:32.245
Um, except that the, the market agents,

259
00:13:32.435 --> 00:13:35.485
they may change their policies, their investment policies,

260
00:13:36.145 --> 00:13:39.325
at least to some extent apart from the signaling, we can

261
00:13:39.505 --> 00:13:41.845
of course, try to improve this kind of indicators.

262
00:13:41.865 --> 00:13:45.005
So we, there are, there are ways even to improve the,

263
00:13:45.465 --> 00:13:50.245
the trending method, but also, maybe, ways how

264
00:13:50.245 --> 00:13:52.405
to improve the, the CA model

265
00:13:52.505 --> 00:13:54.965
or the, the general modeling around.

266
00:13:56.105 --> 00:13:59.765
So there's a, a lot of things to do also in particular

267
00:13:59.785 --> 00:14:04.485
to combine our purely statistical method here, maybe

268
00:14:04.515 --> 00:14:05.805
with some, some theory

269
00:14:06.035 --> 00:14:08.485
because, we need some theory

270
00:14:08.585 --> 00:14:09.885
to understand better what's,

271
00:14:09.885 --> 00:14:11.605
what's going on there and what's happening there.

272
00:14:12.105 --> 00:14:16.405
So we may try to find, some kind of bubble theories

273
00:14:16.465 --> 00:14:21.205
or market, theories which, in include these

274
00:14:21.755 --> 00:14:25.605
kind of situations in which we may have unusual conditions

275
00:14:25.605 --> 00:14:26.685
and unusual movements.

Exploring the Predictability of Financial Market Bubbles

00:00:00 - We had the question why nobody really can identify or even predict, what we sometimes perceive as a, as a bubble or pressure in financial markets. so therefore we, we couldn't really understand why, why that's so difficult. And therefore, we started to think about that and try to find out if there we can anything, do in order to, get an idea what happens there. And, maybe, that it is even early to detect.

00:00:34 - and, and knew about the bubble literature, but there's no really consensus in, in theory what the bubble actually is and what the crash is. And, so the only thing which people find is, and, that's also what we, what we see in public, that at the end with the crash, there's a, there's a huge damage, but what the bubble is and how we get into the bubble and, if it is predictable even, is something which, where there's no really theory and no really, clear consensus about that.

00:01:08 - to look at purely at data and then try to analyze the data, and maybe we find out something about the data. And therefore, the question we had is, is it possible to identify normal market behavior and something like a normal market behavior? So normal or abnormal was the question here. or is it possible even to predict early enough that resist some, some kind of future period, maybe a year?

Analyzing Market Behavior: Normal vs. Anomalous Conditions

00:01:38 - an anomalous condition, and that is something we may identify purely statistically. So then the question is, what is normal? What is un-normal, of course, or what is a normal unnormal behavior? And again, we go to pure statistics. And what you do in, time series analysis is that you start to decompose your, your time series. And the time series we looked at here was the

00:02:07 - And, here we looked at the earning to price ratios, AP ratios as, one of the standard indicators that may give us some idea about, the market conditions. it was very simple actually, what we could define as normal or un-normal. so as known and even well known from something like the Bollinger Bands, we identify the trend movements and then defined a band around these trend movements

00:02:41 - that might be in this band. And then the economic interpretation is that the trend is something like the fundamental market movement and, that there is some, some kind of other systematic or random movement around, we identified that as long the market is within This band, we see that we are under kind of normal behavior of the market under normal conditions. however, if the market, is moving out of the bands,

Innovative Statistical Methods for Trend Identification

00:03:13 - And that is something we can purely look, up from the statistics and by statistic definitions. So up to that point, nothing is really new. However, the novelty here and now comes when we look at the methods, how we do the decomposition, and, let's start maybe with a, with a, with a trend. So we look at the trend and for the trend, and the identification of the trend, we introduce an endogenous local linear

00:03:44 - This is innovative, because it's fully data driven and doesn't need, some, some kind of arbitrary parameters, which we have to put in. And, these kind of methodology actually was developed earlier in several papers, in a group of researchers at this university, namely Huang and Marlon Fit, and myself and some others were involved in this, development of, of, of this methodology of this. And ous,

00:04:16 - and, the trend can be identified easily, better than, some, some other trend methods, because we, again, use statistical criteria in order to, get to the, the optimal, determination of, of this trend. So it is, a method which is better than the often used HP filter method or some other method. Therefore, the, this new,

Analyzing Residuals for Predictive Market Insights

00:04:48 - of our analysis. so the second, methodology we used, after detecting the trend, we looked at the residuals and then tried to find out is there something systematic in the residuals? And, actually we did that with Aceta model and these kind of C models, really could identify, systematic movements apart from the trend.

00:05:22 - model good prediction models in which we, now in the third step could do predictions, both for the normal regions and the development of the normal region of the market, but also for the abnormalities and for also the, the market itself. So with the help of, the CER and these, local linear trend methodology,

00:05:51 - we had some novelties in our methodology and that, probably helped to get the good Performance of the indicator at the end. CTA models is quite important. CTA model is an auto aggressive model. so what we do is we identify, that there is in the residual. So after the det trending method, we look at the residuals and the residuals actually, they have some systematic movements and these systematic systematic movements,

Asymmetric Movements in Market Predictions

00:06:23 - And, therefore with the CT a model, we also get the improvement that the, direction, is, can be asymmetric. So the movements upwards may be different than the movements downwards. and, for each direction, for the upwards and downward movements, we get a, an extra kind of modeling. and that can be used also

00:06:52 - So therefore, these, the CA model, goes in both directions and identified the systematic market movements in both directions, even if they are sym asymmetric. This asymetry is something which we also identified, and we identified the systematic movement. And this systematic movement could be used later on, for the predictions. and the indicator, what we did not want to do is we did not want to predict that,

00:07:24 - so that is nothing, which is predictable actually. however, what we can predict maybe is that the market is on the way to turn animals and, this may happen within the next year. And then once the market moves in an animal region, then the chance for turning point or market correction is clearly increasing. And so therefore, also it, what is increasing is

Identifying Anomalies in Financial Markets

00:07:57 - the market may even turn into something like a, a bubble or crash at the end, there may be a correction and a turning point ahead, and therefore, the first step is to identify when, market turns anomalous. And, this is what, what we try to do. And again, also with a, with a specification, with a choice of the specification of the CTA model, we use purely statistical criteria. And, with the help of these criteria,

00:08:30 - if there are anomalies ahead. and the, the rate, how we identified the quality, the capability of, detecting that is by the two positive rate and the two positive rate. here was 80, 80%. So 80% of all the exposed with all information Existing anomalies, were detected, by,

00:09:00 - The question is, how and what could we detect there? however, first, what, what it, what it means, if we, I identify something like 23, anomalies episodes, it means that from time to time, the market moves into something unusual. So it's a simply statistical idea. It is something which is unusual. However, this may not be a problem. So the market may correct somehow and turn back to, to normal after this unusual episode.

Identifying Anomalous Market Conditions and Risks

00:09:36 - however, there may be also something, as a result of this episode, which at the end is harmful. So normally it's, not really harmful, but, in some cases, these anom, anomalous conditions, turn out to be something which we later perceive as bubbles and crashes. So therefore, detecting anomalous episodes is,

00:10:08 - is not really a big thing. however, maybe it turns into a bubble in a crash, and therefore it makes sense to give a signal in the market a warning signal in the market here, something which is anomalous and therefore maybe there's a crash ahead. Therefore, you view the market, the market agents, should, become a little bit more careful in your decisions and, and, and check if you really wanna stay in that market.

00:10:40 - therefore, what we discovered here is, 70 or 80% of these 23 anomalous conditions. it started in 1970 to 2020, in 1970 because, we needed the time from, before 1970, after the second World War, in order to adjust the model, the model needs also some, some time before, however, afterwards, with the two positive rate,

Predicting Market Anomalies: A Warning Signal for Bubbles and Crashes

00:11:13 - And, now the idea is that, the indicator may, may signal in the market, yes, there's something unusual and maybe even something like a bubble ahead. And the good thing is that with this method, we could identify these kind of anomalous episodes that at the end turned out to be bubbles like the bubble and the crash in 1987. we could, identify this, one year ahead

00:11:47 - and crash and overvaluation. so if these, indicator were usable before, at least a year ahead or something, we could have given signals in the market so that, something unusual is going on. And with this, something unusual going on, the market could have decided to turn back to normal.

00:12:18 - the crashes and the bubbles. however that, that is the idea, a warning signal in order to give warnings. And, and, there's something like an market correction from the market itself. We do not, think it makes a big sense that, some kind of agency should step in and do something if the signal comes up. however, if if the market realizes something,

Assessing Market Anomalies and Decision-Making

00:12:50 - is it unusual because, it should be unusual at this point, or is it unusual? And the market should, be more carefully think about, what, what decisions they make, if they wanna stay and, and right maybe on the, on the market movements, or should they really, be more careful and, and, check if they are still on the, on the right side.

00:13:21 - to give an orientation. but there's no really particular policy involved in that. except that the, the market agents, they may change their policies, their investment policies, at least to some extent apart from the signaling, we can of course, try to improve this kind of indicators. So we, there are, there are ways even to improve the, the trending method, but also, maybe, ways how

00:13:52 - or the, the general modeling around. So there's a, a lot of things to do also in particular to combine our purely statistical method here, maybe with some, some theory because, we need some theory to understand better what's, what's going on there and what's happening there. So we may try to find, some kind of bubble theories or market, theories which, in include these

Analyzing Unusual Market Movements

00:14:25 - and unusual movements.

Faculti AI can make mistakes

Research

Policy

Analysis

Data

Originals

Explainers

Podcasts

Faculti

New on Faculti

Idea Explorer

Definition

News Context

Critical Questions and Answers

Research Summaries

An early indicator for anomalous stock market performance

Media Literacy Toolkit

Critical Questions to Consider ▼

Bias and Perspective Awareness ▼

Further Reading ▼