WEBVTT

1
00:00:00.020 --> 00:00:00.200
Hello,

2
00:00:00.340 --> 00:00:02.320
welcome back to Papers with Backtest podcast.

3
00:00:02.701 --> 00:00:02.961
Today,

4
00:00:03.061 --> 00:00:05.441
we dive into another algo trading research paper.

5
00:00:05.781 --> 00:00:06.180
Indeed.

6
00:00:06.281 --> 00:00:08.703
And this one looks at something maybe a bit different,

7
00:00:09.445 --> 00:00:11.086
overnight stock returns.

8
00:00:11.164 --> 00:00:11.602
Exactly.

9
00:00:11.742 --> 00:00:15.086
The return from the close one day to the open the next.

10
00:00:15.344 --> 00:00:16.203
And the big question is,

11
00:00:16.406 --> 00:00:20.688
can this tell us something about how investors specifically feel about a company?

12
00:00:21.078 --> 00:00:21.203
Yeah,

13
00:00:21.204 --> 00:00:21.984
the core idea,

14
00:00:22.078 --> 00:00:23.500
and it builds on some previous work,

15
00:00:24.109 --> 00:00:26.703
is that maybe individual investor sentiment,

16
00:00:27.500 --> 00:00:27.750
they're...

17
00:00:28.048 --> 00:00:33.749
optimism or pessimism shows up in that after hours trading and gets baked into the overnight return.

18
00:00:34.089 --> 00:00:37.452
It's an interesting angle trying to capture sentiment at the firm level,

19
00:00:37.491 --> 00:00:37.710
right?

20
00:00:37.827 --> 00:00:39.351
Not just broad market mood.

21
00:00:39.609 --> 00:00:39.812
Right.

22
00:00:39.827 --> 00:00:40.507
So this paper,

23
00:00:40.929 --> 00:00:44.671
it's called Overnight Returns and Firm-Specific Investor Sentiment.

24
00:00:44.952 --> 00:00:47.515
And they're basically testing if this overnight return measure,

25
00:00:47.827 --> 00:00:47.999
well,

26
00:00:48.077 --> 00:00:49.890
if it actually acts like a sentiment indicator.

27
00:00:49.968 --> 00:00:50.468
Makes sense.

28
00:00:50.687 --> 00:00:54.046
They needed to see if its characteristics line up with what you'd expect from sentiment.

29
00:00:54.249 --> 00:00:55.656
So they use CRSP data.

30
00:00:56.040 --> 00:01:00.599
Looked at the periods from July 1992 through December 2013.

31
00:01:00.600 --> 00:01:04.783
And 92 is key because that's when reliable open price data became available.

32
00:01:05.103 --> 00:01:05.463
Got it.

33
00:01:05.986 --> 00:01:07.064
And for our deep dive today,

34
00:01:07.103 --> 00:01:13.338
we're really going to focus on the potential trading rules and the backtest results they found.

35
00:01:13.384 --> 00:01:13.509
OK,

36
00:01:13.681 --> 00:01:15.963
so the paper breaks down into a few main areas.

37
00:01:16.650 --> 00:01:17.009
First,

38
00:01:17.338 --> 00:01:19.244
does this overnight return persist?

39
00:01:19.619 --> 00:01:19.869
You know,

40
00:01:19.884 --> 00:01:22.681
does a high return follow a high return in the short term?

41
00:01:23.228 --> 00:01:25.610
Then how does that link up with the type of company?

42
00:01:25.750 --> 00:01:25.930
Like,

43
00:01:25.971 --> 00:01:27.754
is it harder to value who owns it?

44
00:01:27.914 --> 00:01:28.813
Lots of institutions.

45
00:01:28.875 --> 00:01:29.035
Okay,

46
00:01:29.075 --> 00:01:30.594
persistence and company characteristics.

47
00:01:30.856 --> 00:01:31.317
And finally,

48
00:01:31.496 --> 00:01:32.739
what happens longer term?

49
00:01:33.356 --> 00:01:35.598
If a stock has really high overnight returns for a bit,

50
00:01:35.684 --> 00:01:37.043
does that reverse later on?

51
00:01:37.285 --> 00:01:38.543
Contrarian possibilities there.

52
00:01:39.004 --> 00:01:39.192
Okay,

53
00:01:39.379 --> 00:01:41.160
let's start with that short-term persistence then.

54
00:01:41.161 --> 00:01:41.410
Right.

55
00:01:41.489 --> 00:01:42.270
How did they test it?

56
00:01:42.660 --> 00:01:44.473
They used a weekly sorting method.

57
00:01:44.567 --> 00:01:45.692
So every single week,

58
00:01:45.864 --> 00:01:49.801
they'd rank all the stocks based on their total overnight return for that week.

59
00:01:50.020 --> 00:01:50.754
Lowest to highest.

60
00:01:50.832 --> 00:01:51.254
Exactly.

61
00:01:51.770 --> 00:01:51.973
And...

62
00:01:52.208 --> 00:01:53.930
divided them into 10 groups,

63
00:01:54.171 --> 00:01:54.730
deciles.

64
00:01:55.050 --> 00:01:55.851
And what popped out?

65
00:01:56.253 --> 00:01:56.370
Well,

66
00:01:56.472 --> 00:01:59.917
something quite noticeable for anyone looking at short-term moves,

67
00:02:00.378 --> 00:02:02.034
stops that were in the top 10%

68
00:02:02.097 --> 00:02:03.558
for overnight returns one week.

69
00:02:03.800 --> 00:02:04.737
The highest performers.

70
00:02:05.159 --> 00:02:05.323
Yeah.

71
00:02:06.003 --> 00:02:13.065
They tended to have a significantly better average overnight return the next week compared to the stocks in the bottom 10%.

72
00:02:13.284 --> 00:02:13.831
How much better?

73
00:02:14.206 --> 00:02:19.472
The difference was about 1.76 percentage points on average for that following week.

74
00:02:19.659 --> 00:02:19.831
Wow.

75
00:02:20.088 --> 00:02:20.208
OK,

76
00:02:20.489 --> 00:02:21.850
1.76%

77
00:02:21.851 --> 00:02:23.652
in a week is definitely not trivial.

78
00:02:23.851 --> 00:02:23.972
No,

79
00:02:24.073 --> 00:02:24.632
it's not.

80
00:02:24.812 --> 00:02:26.855
And you said it wasn't just a one-week phenomenon.

81
00:02:27.355 --> 00:02:28.273
It kind of lingered.

82
00:02:28.515 --> 00:02:29.437
It did seem to.

83
00:02:29.898 --> 00:02:32.679
The effect diminished week by week,

84
00:02:32.843 --> 00:02:36.202
but it was still statistically significant for up to four weeks later.

85
00:02:36.405 --> 00:02:37.390
So W plus 1,

86
00:02:37.968 --> 00:02:39.265
W plus 2,

87
00:02:39.280 --> 00:02:40.421
up to W plus 4.

88
00:02:40.609 --> 00:02:41.109
That's right.

89
00:02:41.562 --> 00:02:47.280
So a big overnight jump seemed to signal a higher probability of more positive overnight returns,

90
00:02:47.312 --> 00:02:48.202
maybe smaller ones.

91
00:02:48.676 --> 00:02:49.657
for the next month or so.

92
00:02:50.097 --> 00:02:50.458
Interesting.

93
00:02:50.738 --> 00:02:55.062
And did the average return generally go up across those 10 groups in the follow-up weeks?

94
00:02:55.323 --> 00:02:55.542
Yes.

95
00:02:55.605 --> 00:03:01.691
They saw that generally the subsequent week's average overnight return increased as you move from the lowest initial decile to the highest.

96
00:03:01.730 --> 00:03:01.847
Now,

97
00:03:01.886 --> 00:03:02.152
hold on.

98
00:03:02.292 --> 00:03:04.152
Could this just be like market mechanics,

99
00:03:04.839 --> 00:03:06.089
bid-ask-bounce or something?

100
00:03:06.261 --> 00:03:07.402
That's a fair question.

101
00:03:07.448 --> 00:03:08.917
The researchers thought about that too.

102
00:03:09.480 --> 00:03:11.402
They re-ran the numbers using,

103
00:03:11.495 --> 00:03:11.792
quote,

104
00:03:12.183 --> 00:03:12.495
data,

105
00:03:13.073 --> 00:03:16.464
specifically the midpoint between the bid and ask prices.

106
00:03:16.620 --> 00:03:18.011
To kind of smooth out the spread effect.

107
00:03:18.136 --> 00:03:18.657
Exactly.

108
00:03:18.917 --> 00:03:20.719
And they found very similar results.

109
00:03:21.078 --> 00:03:27.766
So it strongly suggests that the bid-ask spread isn't the main thing driving this short-term continuation.

110
00:03:28.008 --> 00:03:28.227
Okay,

111
00:03:28.328 --> 00:03:29.289
that's a crucial check.

112
00:03:29.750 --> 00:03:37.782
So how does this overnight persistence compare to just looking at regular close-to-close returns over the same weekly period?

113
00:03:38.016 --> 00:03:38.235
Ah,

114
00:03:38.391 --> 00:03:39.235
good comparison.

115
00:03:39.516 --> 00:03:43.297
When they did the same decile sorting based on weekly close-to-close returns,

116
00:03:43.657 --> 00:03:45.282
the picture was much fuzzier.

117
00:03:45.547 --> 00:03:46.282
No clear pattern.

118
00:03:46.356 --> 00:03:46.736
Pretty much.

119
00:03:46.877 --> 00:03:50.719
No consistent monotonic increase across the deciles in the next week.

120
00:03:51.102 --> 00:03:54.524
And the differences between the top and bottom groups were smaller and,

121
00:03:54.688 --> 00:03:55.024
frankly,

122
00:03:55.281 --> 00:03:55.984
less reliable.

123
00:03:56.102 --> 00:04:00.227
So it really points to something specific happening in that overnight close to open window.

124
00:04:00.391 --> 00:04:01.227
It seems that way,

125
00:04:01.328 --> 00:04:01.508
yeah.

126
00:04:01.633 --> 00:04:05.461
Like that period is particularly sensitive to whatever's causing this persistence.

127
00:04:05.462 --> 00:04:06.539
They did robustness checks,

128
00:04:06.570 --> 00:04:06.774
right?

129
00:04:06.820 --> 00:04:10.055
Made sure it wasn't just size or momentum or something else explaining it away.

130
00:04:10.367 --> 00:04:10.477
Oh,

131
00:04:10.586 --> 00:04:11.024
definitely.

132
00:04:11.117 --> 00:04:13.055
They controlled for standard factors.

133
00:04:13.586 --> 00:04:14.164
Market beta,

134
00:04:14.399 --> 00:04:15.070
from size,

135
00:04:15.211 --> 00:04:15.867
book to market.

136
00:04:16.088 --> 00:04:16.688
momentum,

137
00:04:17.368 --> 00:04:18.728
the persistence held up.

138
00:04:19.070 --> 00:04:21.109
Even when they looked within different groups,

139
00:04:21.167 --> 00:04:21.308
say,

140
00:04:21.390 --> 00:04:24.991
just within large cap stocks or just within high momentum stocks,

141
00:04:25.030 --> 00:04:30.351
they still found that stocks with higher overnight returns tended to have higher overnight returns the next week.

142
00:04:30.648 --> 00:04:30.874
Okay.

143
00:04:31.070 --> 00:04:33.437
That makes the short-term findings seem pretty solid then.

144
00:04:33.546 --> 00:04:35.390
There's some kind of continuation happening there.

145
00:04:35.468 --> 00:04:36.015
Seems like it.

146
00:04:36.171 --> 00:04:36.405
All right.

147
00:04:36.406 --> 00:04:39.077
So let's connect this to those firm characteristics you mentioned.

148
00:04:39.577 --> 00:04:41.468
This could refine potential trading rules.

149
00:04:41.530 --> 00:04:41.827
First,

150
00:04:42.296 --> 00:04:43.749
the hard-to-value stocks.

151
00:04:43.999 --> 00:04:45.577
How did they define hard to value?

152
00:04:45.776 --> 00:04:47.596
They used several proxies,

153
00:04:47.997 --> 00:04:51.676
things that usually indicate more uncertainty or less available information.

154
00:04:51.958 --> 00:04:52.278
Like what?

155
00:04:53.036 --> 00:04:54.219
Stock return volatility,

156
00:04:54.337 --> 00:04:55.094
more volatile,

157
00:04:55.337 --> 00:04:56.094
harder to value.

158
00:04:56.298 --> 00:04:57.001
Firm size,

159
00:04:57.118 --> 00:04:58.555
smaller firms are often trickier.

160
00:04:59.055 --> 00:05:00.782
Firm age younger means less history.

161
00:05:01.353 --> 00:05:01.974
Profitability.

162
00:05:02.935 --> 00:05:04.576
Less profitable can mean more uncertainty.

163
00:05:04.957 --> 00:05:05.076
OK.

164
00:05:05.398 --> 00:05:07.078
And also the earnings to price ratio.

165
00:05:07.898 --> 00:05:10.941
A low EP often implies high growth expectations,

166
00:05:10.965 --> 00:05:13.246
which are inherently harder to value accurately.

167
00:05:13.645 --> 00:05:14.066
Makes sense.

168
00:05:14.668 --> 00:05:16.809
So five different ways to slice difficulty.

169
00:05:17.324 --> 00:05:19.949
How did they relate this back to the overnight returns?

170
00:05:20.230 --> 00:05:21.340
They did a two step sort.

171
00:05:21.855 --> 00:05:22.246
First,

172
00:05:22.418 --> 00:05:27.137
each year they'd group stocks into quartiles based on one of those hard to value metrics,

173
00:05:27.293 --> 00:05:28.762
say volatility.

174
00:05:29.199 --> 00:05:30.418
four groups from least

175
00:05:30.565 --> 00:05:31.346
to most volatile.

176
00:05:31.446 --> 00:05:31.946
Exactly.

177
00:05:32.527 --> 00:05:34.971
Then within each of those four volatility groups,

178
00:05:35.229 --> 00:05:38.033
they did the weekly decile sort based on overnight returns,

179
00:05:38.213 --> 00:05:38.912
just like before.

180
00:05:39.131 --> 00:05:39.397
I see.

181
00:05:39.451 --> 00:05:47.201
So sorting within sorts and what emerged was the persistence stronger or weaker for the hard to value ones?

182
00:05:47.264 --> 00:05:47.701
Stronger,

183
00:05:47.936 --> 00:05:48.967
significantly stronger.

184
00:05:49.123 --> 00:05:52.842
And this was consistent across all five of their hard to value proxies.

185
00:05:52.904 --> 00:05:53.123
Really?

186
00:05:53.373 --> 00:05:53.654
For all

187
00:05:54.029 --> 00:05:56.170
Yeah. The difference in the next week's...

188
00:05:56.697 --> 00:06:04.917
W plus one overnight return between the top and bottom overnight deciles was always biggest in that quartile representing the most difficult to value stocks.

189
00:06:04.995 --> 00:06:07.362
Can you give an example like with volatility or size?

190
00:06:07.479 --> 00:06:07.659
Sure.

191
00:06:08.042 --> 00:06:08.776
For volatility,

192
00:06:08.862 --> 00:06:19.839
the top minus bottom decile difference in next week's overnight return was one point nine nine percentage points for the most volatile quartile compared to only one point zero four percentage points for the least volatile.

193
00:06:19.840 --> 00:06:20.526
It's almost double.

194
00:06:20.636 --> 00:06:20.995
It is.

195
00:06:21.229 --> 00:06:21.948
And for size,

196
00:06:21.949 --> 00:06:22.995
it was even more stark.

197
00:06:23.625 --> 00:06:26.428
2.32 percentage points for the smallest,

198
00:06:26.807 --> 00:06:31.893
hardest to value quartile versus just 0.72 percentage points for the largest firms.

199
00:06:32.354 --> 00:06:32.713
Wow.

200
00:06:33.596 --> 00:06:33.838
Okay.

201
00:06:33.877 --> 00:06:41.862
So the implication here for a trader might be focus short-term sentiment strategies on these harder to pin down stocks.

202
00:06:42.190 --> 00:06:43.846
That's certainly what the results suggest.

203
00:06:44.377 --> 00:06:49.252
Sentiment seems to pack a bigger punch in the short run when fundamental value is more ambiguous.

204
00:06:49.502 --> 00:06:49.784
Okay.

205
00:06:49.877 --> 00:06:51.237
Now the other characteristic.

206
00:06:52.293 --> 00:06:53.334
Institutional ownership.

207
00:06:53.374 --> 00:06:54.376
What was the idea there?

208
00:06:54.614 --> 00:06:56.876
The hypothesis was that sentiment,

209
00:06:57.400 --> 00:07:00.122
especially the kind potentially driving overnight returns,

210
00:07:00.501 --> 00:07:04.282
is more associated with individual investors rather than large institutions.

211
00:07:04.368 --> 00:07:07.111
Because institutions are maybe more fundamentals driven,

212
00:07:07.267 --> 00:07:09.251
less swayed by short term noise?

213
00:07:09.345 --> 00:07:10.251
That's the general thinking.

214
00:07:10.329 --> 00:07:15.157
So they predicted the overnight persistence effect would be weaker in stocks with high institutional ownership.

215
00:07:15.158 --> 00:07:16.454
And did they test that the same way?

216
00:07:16.751 --> 00:07:18.251
Sorting by ownership level first?

217
00:07:18.564 --> 00:07:18.767
Yes,

218
00:07:18.923 --> 00:07:19.626
similar approach.

219
00:07:19.837 --> 00:07:25.164
They sorted stocks into four groups based on the percentage of shares held by institutions from lowest to highest I.O.

220
00:07:25.664 --> 00:07:29.230
And then the weekly overnight return deciles within each I.O.

221
00:07:29.324 --> 00:07:29.566
group.

222
00:07:29.683 --> 00:07:30.230
Precisely.

223
00:07:30.425 --> 00:07:31.050
And the result.

224
00:07:31.550 --> 00:07:32.269
Did high I.O.

225
00:07:32.832 --> 00:07:33.652
dampen the effect?

226
00:07:33.949 --> 00:07:34.394
It did.

227
00:07:34.613 --> 00:07:35.550
The persistence,

228
00:07:35.613 --> 00:07:39.355
that next week return difference between the top and bottom overnight deciles,

229
00:07:39.793 --> 00:07:43.714
it systematically decreased as institutional ownership increased.

230
00:07:43.824 --> 00:07:44.636
How much of a decrease?

231
00:07:44.839 --> 00:07:45.214
For Week

232
00:07:45.589 --> 00:07:46.339
W Plus One.

233
00:07:46.553 --> 00:07:50.599
The difference was 2.36 percentage points for the lowest IO quartile,

234
00:07:50.919 --> 00:07:54.982
but it fell to 1.07 percentage points for the highest IO quartile.

235
00:07:55.060 --> 00:07:55.302
Again,

236
00:07:55.380 --> 00:07:56.341
quite a significant drop.

237
00:07:56.583 --> 00:07:57.380
More than halved.

238
00:07:57.560 --> 00:07:57.747
Yeah.

239
00:07:58.021 --> 00:08:06.388
It really supports the idea that this overnight phenomenon is more strongly linked to segments of the market where institutions aren't the dominant players.

240
00:08:07.013 --> 00:08:09.982
So another potential filter for a short-term strategy.

241
00:08:10.749 --> 00:08:15.493
Perhaps favor stocks with lower institutional holdings if you're playing this sentiment persistence.

242
00:08:15.552 --> 00:08:17.575
That could be a logical conclusion from these findings,

243
00:08:17.614 --> 00:08:17.794
yes.

244
00:08:17.935 --> 00:08:18.415
High I.O.

245
00:08:18.536 --> 00:08:21.122
stocks seem less susceptible to this particular effect.

246
00:08:21.317 --> 00:08:21.481
Okay,

247
00:08:21.521 --> 00:08:24.021
that covers the short-term and how firm type matters.

248
00:08:24.724 --> 00:08:25.864
What about the flip side,

249
00:08:26.067 --> 00:08:26.950
the longer-term picture?

250
00:08:27.044 --> 00:08:28.325
You mentioned potential reversals.

251
00:08:28.810 --> 00:08:29.044
Right.

252
00:08:29.481 --> 00:08:31.247
So does excessive sentiment,

253
00:08:31.310 --> 00:08:33.778
as may be proxied by these overnight returns,

254
00:08:34.169 --> 00:08:36.075
lead to longer-term corrections?

255
00:08:36.138 --> 00:08:37.372
A classic sentiment story.

256
00:08:37.653 --> 00:08:38.810
How did they investigate this?

257
00:08:39.085 --> 00:08:40.486
They switched gears a bit.

258
00:08:40.607 --> 00:08:41.748
Instead of weekly source,

259
00:08:41.767 --> 00:08:43.449
they formed portfolios monthly,

260
00:08:43.691 --> 00:08:45.230
specifically every December.

261
00:08:45.390 --> 00:08:46.011
Why December?

262
00:08:46.691 --> 00:08:49.492
Likely just to have a consistent annual rebalancing point.

263
00:08:49.914 --> 00:08:55.179
They ranked stocks based on their average daily overnight return over that entire month of December.

264
00:08:55.336 --> 00:08:55.523
Okay.

265
00:08:55.679 --> 00:08:56.586
Average for the whole month?

266
00:08:57.039 --> 00:08:57.882
Then DeSiles again.

267
00:08:57.929 --> 00:08:58.523
DeSiles again.

268
00:08:58.601 --> 00:09:00.039
Lowest average to highest average.

269
00:09:00.070 --> 00:09:00.711
And the strategy?

270
00:09:00.742 --> 00:09:01.101
A long,

271
00:09:01.179 --> 00:09:01.742
short approach.

272
00:09:02.333 --> 00:09:06.636
Go long the bottom decile stocks with the lowest average overnight returns that month,

273
00:09:06.737 --> 00:09:08.901
suggesting maybe pessimistic sentiment.

274
00:09:09.542 --> 00:09:13.167
And go short the top decile stocks with the highest average returns,

275
00:09:13.222 --> 00:09:14.886
maybe overly optimistic sentiment.

276
00:09:14.964 --> 00:09:16.003
And hold for how long?

277
00:09:16.245 --> 00:09:19.175
They held these long short portfolios for the next 12 months.

278
00:09:19.347 --> 00:09:19.487
OK,

279
00:09:19.972 --> 00:09:25.159
a longer term contrarian bet against the prior month's extreme overnight movers.

280
00:09:25.800 --> 00:09:27.300
What did the back test show?

281
00:09:27.690 --> 00:09:30.300
It showed a significant positive abnormal return.

282
00:09:30.981 --> 00:09:32.783
After adjusting for the usual risk factors,

283
00:09:32.882 --> 00:09:33.263
market,

284
00:09:33.363 --> 00:09:33.804
size,

285
00:09:33.923 --> 00:09:34.324
value,

286
00:09:34.365 --> 00:09:34.804
momentum,

287
00:09:35.267 --> 00:09:36.728
they used a five-factor model.

288
00:09:37.209 --> 00:09:41.455
The strategy generated an average monthly alpha of 0.62 percentage points.

289
00:09:41.548 --> 00:09:42.994
0.62%

290
00:09:42.995 --> 00:09:43.275
per month.

291
00:09:43.276 --> 00:09:43.634
That's true.

292
00:09:43.916 --> 00:09:44.056
Yeah.

293
00:09:44.134 --> 00:09:44.798
That's substantial.

294
00:09:44.853 --> 00:09:45.392
It adds up.

295
00:09:45.572 --> 00:09:45.955
Annually,

296
00:09:46.017 --> 00:09:47.814
that's around 7.4%

297
00:09:47.815 --> 00:09:48.033
alpha.

298
00:09:48.377 --> 00:09:48.736
Wow.

299
00:09:49.314 --> 00:09:49.439
So

300
00:09:49.882 --> 00:09:52.843
It really suggests that stocks getting pushed up hard overnight,

301
00:09:52.923 --> 00:09:54.063
perhaps on sentiment,

302
00:09:54.481 --> 00:09:57.665
tend to underperform significantly over the following year and vice versa.

303
00:09:57.985 --> 00:09:58.485
Exactly.

304
00:09:58.743 --> 00:10:03.305
It provides evidence for a longer term reversal based on this overnight sentiment proxy,

305
00:10:03.462 --> 00:10:05.305
a potential contrarian strategy.

306
00:10:05.563 --> 00:10:09.813
And did this reversal effect also vary with how hard the stocks were to value?

307
00:10:10.032 --> 00:10:11.891
Did sentiment matter more there in the long run,

308
00:10:11.923 --> 00:10:12.063
too?

309
00:10:12.391 --> 00:10:12.641
Yes.

310
00:10:12.923 --> 00:10:13.360
Interestingly,

311
00:10:13.361 --> 00:10:13.704
it did.

312
00:10:13.813 --> 00:10:18.813
When they looked at the performance of this long short strategy within those different hard to value subgroups.

313
00:10:18.829 --> 00:10:18.985
Uh-huh.

314
00:10:19.442 --> 00:10:26.351
The positive abnormal returns were generally stronger and more consistently significant for the stocks deemed most difficult to value.

315
00:10:26.507 --> 00:10:29.194
So not only is the short-term persistence stronger in those stocks,

316
00:10:29.210 --> 00:10:31.554
but the eventual reversal seems more pronounced,

317
00:10:31.577 --> 00:10:31.757
too.

318
00:10:32.093 --> 00:10:33.359
That's what the data indicated.

319
00:10:33.976 --> 00:10:36.866
It reinforces the idea that sentiment effects,

320
00:10:37.116 --> 00:10:39.819
both the initial momentum and the subsequent correction,

321
00:10:40.319 --> 00:10:43.038
are amplified when fundamentals are less certain.

322
00:10:43.382 --> 00:10:43.632
Okay.

323
00:10:43.819 --> 00:10:46.523
They also briefly touched on earnings announcements,

324
00:10:46.585 --> 00:10:46.944
didn't they?

325
00:10:47.302 --> 00:10:49.425
Just as another way to show this measure matters.

326
00:10:49.525 --> 00:10:49.724
Yeah,

327
00:10:49.804 --> 00:10:50.726
it was more illustrative.

328
00:10:50.745 --> 00:10:54.511
They found that the level of pre-announcement overnight returns,

329
00:10:54.570 --> 00:10:55.031
basically,

330
00:10:55.527 --> 00:11:00.753
the sentiment leading into the announcement affected how the price reacted to the earnings news itself.

331
00:11:01.019 --> 00:11:01.394
How so?

332
00:11:01.816 --> 00:11:02.597
If sentiment,

333
00:11:03.019 --> 00:11:04.113
high overnight returns,

334
00:11:04.566 --> 00:11:06.597
was already optimistic before the announcement,

335
00:11:06.956 --> 00:11:11.347
the positive price reaction to the actual earnings report tended to be weaker.

336
00:11:11.613 --> 00:11:11.972
It's like...

337
00:11:12.282 --> 00:11:17.467
Some of the good news or perhaps excessive optimism was already priced in via that overnight sentiment.

338
00:11:17.487 --> 00:11:17.705
I see.

339
00:11:17.729 --> 00:11:20.229
So the overnight return isn't just predicting future returns.

340
00:11:20.549 --> 00:11:23.455
It's also conditioning how the market reacts to new information.

341
00:11:23.737 --> 00:11:24.213
Exactly.

342
00:11:24.455 --> 00:11:29.823
It demonstrates that this firm specific sentiment measure has tangible impacts on market dynamics.

343
00:11:29.901 --> 00:11:30.026
OK,

344
00:11:30.198 --> 00:11:31.963
let's try to summarize the key takeaways here,

345
00:11:32.135 --> 00:11:33.620
especially for algo traders listening.

346
00:11:34.041 --> 00:11:35.463
What should they be thinking about?

347
00:11:35.760 --> 00:11:35.885
Well,

348
00:11:35.948 --> 00:11:36.229
first,

349
00:11:36.307 --> 00:11:38.495
these overnight returns aren't just random noise.

350
00:11:38.791 --> 00:11:40.416
They show short term persistence.

351
00:11:40.866 --> 00:11:42.287
especially over the next week or so.

352
00:11:42.527 --> 00:11:43.969
Potential momentum signals there.

353
00:11:44.170 --> 00:11:44.430
Right.

354
00:11:45.051 --> 00:11:45.430
Second,

355
00:11:45.691 --> 00:11:49.094
this persistence seems stronger in specific types of stocks,

356
00:11:49.914 --> 00:11:55.320
those that are harder to value based on various metrics and those with lower institutional ownership.

357
00:11:55.758 --> 00:11:59.930
So maybe fertile ground for short-term sentiment strategies in those segments.

358
00:11:59.945 --> 00:12:00.399
Potentially,

359
00:12:00.461 --> 00:12:00.649
yeah.

360
00:12:01.211 --> 00:12:02.367
But then there's the third point,

361
00:12:02.992 --> 00:12:04.414
the longer-term reversal.

362
00:12:05.633 --> 00:12:09.149
High short-term overnight returns seem to predict longer-term

363
00:12:09.338 --> 00:12:10.199
underperformance.

364
00:12:10.279 --> 00:12:13.481
Suggesting contrarian opportunities if you have a longer horizon.

365
00:12:13.563 --> 00:12:14.024
Precisely.

366
00:12:14.242 --> 00:12:19.352
It highlights that tension between short-term sentiment continuation and longer-term mean reversion.

367
00:12:19.906 --> 00:12:20.586
And fundamentally,

368
00:12:20.688 --> 00:12:27.696
the paper makes a case that the overnight return can serve as a useful quantifiable proxy for firm-specific investor sentiment.

369
00:12:28.117 --> 00:12:28.477
It does.

370
00:12:28.478 --> 00:12:32.242
It gives you a number to potentially work with rather than just a vague notion of mood.

371
00:12:32.524 --> 00:12:33.649
So the final thought for you,

372
00:12:33.696 --> 00:12:34.196
the listener,

373
00:12:34.242 --> 00:12:34.539
might be,

374
00:12:35.196 --> 00:12:36.946
how could you incorporate a measure like this?

375
00:12:38.286 --> 00:12:39.808
Thinking about overnight action,

376
00:12:40.128 --> 00:12:44.252
maybe combined with factors like valuation difficulty or institutional presence.

377
00:12:44.471 --> 00:12:44.650
Yeah.

378
00:12:44.771 --> 00:12:48.439
How might it enhance existing strategies or could it form the basis of a new one?

379
00:12:49.158 --> 00:12:53.158
Balancing that short term momentum against the longer term reversal is likely key.

380
00:12:53.580 --> 00:12:54.541
It's definitely food for thought.

381
00:12:54.822 --> 00:12:57.088
Thank you for tuning in to Papers with Backtest podcast.

382
00:12:57.400 --> 00:12:59.572
We hope today's episode gave you useful insights.

383
00:13:00.103 --> 00:13:02.088
Join us next time as we break down more research.

384
00:13:02.588 --> 00:13:03.978
And for more papers and backtests,

385
00:13:04.338 --> 00:13:07.213
find us at https.paperswithbacktest.com.

386
00:13:07.964 --> 00:13:08.597
Happy trading!