WEBVTT

1
00:00:00.020 --> 00:00:00.220
Hello.

2
00:00:00.480 --> 00:00:02.602
Welcome back to Papers with Backtest podcast.

3
00:00:02.982 --> 00:00:03.242
Today,

4
00:00:03.303 --> 00:00:05.926
we dive into another algo trading research paper.

5
00:00:06.105 --> 00:00:06.484
Hi there.

6
00:00:06.828 --> 00:00:06.969
Yeah,

7
00:00:07.008 --> 00:00:09.727
today we're looking at a really interesting one called percent accruals.

8
00:00:10.109 --> 00:00:10.969
It's by Hasala,

9
00:00:11.109 --> 00:00:11.594
Lundholm,

10
00:00:11.734 --> 00:00:13.773
and Van Winkle from 2010.

11
00:00:14.117 --> 00:00:14.336
Right.

12
00:00:14.617 --> 00:00:18.734
And what's neat is how it challenges the sort of standard way we think about accruals,

13
00:00:18.859 --> 00:00:19.719
especially for trading.

14
00:00:20.094 --> 00:00:20.516
Exactly.

15
00:00:20.781 --> 00:00:26.125
It puts forward this idea that maybe the traditional calculation isn't the most effective for spotting opportunities.

16
00:00:26.784 --> 00:00:27.645
For traders listening,

17
00:00:27.665 --> 00:00:29.587
the big question this paper tackles is,

18
00:00:30.647 --> 00:00:34.753
does this new definition they call percent accruals actually work better?

19
00:00:35.190 --> 00:00:39.659
Is it a sharper tool for finding potentially mispriced stocks compared to,

20
00:00:39.956 --> 00:00:40.198
you know,

21
00:00:40.199 --> 00:00:40.675
the old way?

22
00:00:41.081 --> 00:00:41.839
That's the core of it.

23
00:00:42.221 --> 00:00:46.768
And our mission today is really to dig into the trading rules they tested and importantly,

24
00:00:46.784 --> 00:00:48.221
the backtest results they got.

25
00:00:48.768 --> 00:00:49.940
What can we actually use?

26
00:00:50.050 --> 00:00:50.206
Okay,

27
00:00:50.284 --> 00:00:50.753
let's do that.

28
00:00:51.206 --> 00:00:52.596
Before we jump into their new idea,

29
00:00:52.675 --> 00:00:56.034
maybe we should quickly cover the basics of the traditional accrual strategy.

30
00:00:56.276 --> 00:00:57.518
What were people doing before?

31
00:00:57.918 --> 00:00:58.098
Sure.

32
00:00:58.399 --> 00:00:59.739
So the standard definition,

33
00:00:59.799 --> 00:01:00.801
the one most people know,

34
00:01:01.301 --> 00:01:02.801
is you take net income,

35
00:01:03.344 --> 00:01:05.067
subtract cash from operations,

36
00:01:05.403 --> 00:01:09.567
and then you divide that whole thing by the company's average total assets.

37
00:01:10.012 --> 00:01:15.910
It's basically trying to isolate the non-cash part of earnings relative to the company's size.

38
00:01:16.051 --> 00:01:16.457
Exactly.

39
00:01:16.801 --> 00:01:18.739
It's a measure of earnings quality in a sense.

40
00:01:18.801 --> 00:01:20.020
And this wasn't just theoretical,

41
00:01:20.082 --> 00:01:20.285
right?

42
00:01:20.457 --> 00:01:22.723
Sloan's work famously built a strategy on this.

43
00:01:22.785 --> 00:01:22.895
Oh,

44
00:01:22.926 --> 00:01:23.082
yeah.

45
00:01:23.223 --> 00:01:25.364
Sloan's 1996 paper was huge.

46
00:01:25.365 --> 00:01:25.832
it showed.

47
00:01:26.168 --> 00:01:29.010
Pretty significant returns from going long,

48
00:01:29.190 --> 00:01:30.835
low accrual stocks and short,

49
00:01:30.874 --> 00:01:31.913
high accrual stocks.

50
00:01:32.514 --> 00:01:35.475
That really put the accrual anomaly on the map for quant traders.

51
00:01:36.038 --> 00:01:37.780
But like a lot of anomalies,

52
00:01:37.842 --> 00:01:40.358
the initial shine maybe wore off a bit over time,

53
00:01:40.421 --> 00:01:42.280
or at least people started questioning it.

54
00:01:42.624 --> 00:01:43.452
That's fair to say.

55
00:01:43.827 --> 00:01:43.952
Yeah,

56
00:01:44.061 --> 00:01:49.577
later research definitely raised questions about how big that effect really was and if it persisted consistently,

57
00:01:49.999 --> 00:01:53.374
which conveniently leads us right back to this paper and their different approach.

58
00:01:53.639 --> 00:01:54.389
Perfect setup.

59
00:01:55.048 --> 00:01:57.449
So what is this percent accruals measure?

60
00:01:57.568 --> 00:01:59.631
How did Hafzal and his team redefine it?

61
00:02:00.010 --> 00:02:00.189
Okay,

62
00:02:00.428 --> 00:02:02.150
the key difference is the denominator.

63
00:02:03.072 --> 00:02:08.791
They still use net income minus cash from operations in the numerator.

64
00:02:09.533 --> 00:02:12.346
But instead of dividing by average total assets,

65
00:02:13.549 --> 00:02:16.049
they divide by the absolute value of net income.

66
00:02:16.346 --> 00:02:17.314
So net income.

67
00:02:17.674 --> 00:02:17.908
Ah,

68
00:02:18.314 --> 00:02:18.580
okay.

69
00:02:18.908 --> 00:02:20.908
So scaling by earnings itself,

70
00:02:21.049 --> 00:02:22.377
not the asset base.

71
00:02:22.840 --> 00:02:23.741
What's the thinking there?

72
00:02:23.901 --> 00:02:25.083
Why make that change?

73
00:02:25.503 --> 00:02:25.643
Well,

74
00:02:25.683 --> 00:02:28.808
their argument is that this gets closer to the composition of earnings.

75
00:02:29.104 --> 00:02:29.647
It asks,

76
00:02:30.288 --> 00:02:34.511
how much of this reported net income number is actual cash versus accounting adjustments?

77
00:02:34.675 --> 00:02:34.909
Right.

78
00:02:34.995 --> 00:02:39.058
Focusing on the cash versus accrual mix within the earnings figure itself.

79
00:02:39.354 --> 00:02:39.729
Exactly.

80
00:02:39.839 --> 00:02:46.323
And they suggest investors might fixate on that bottom line net income figure without fully distinguishing between the cash and accrual components,

81
00:02:46.683 --> 00:02:48.229
potentially leading to mispricings.

82
00:02:48.542 --> 00:02:48.667
OK.

83
00:02:48.924 --> 00:02:49.925
That makes intuitive sense.

84
00:02:50.025 --> 00:02:51.868
How does this translate into a trading rule then?

85
00:02:51.927 --> 00:02:53.067
How do they sort stocks?

86
00:02:53.270 --> 00:02:55.329
It's a pretty standard portfolio sorting approach.

87
00:02:55.731 --> 00:02:58.716
Firms that have large negative accruals compared to their absolute earnings,

88
00:02:58.731 --> 00:03:01.575
meaning cash flow is much stronger than reported income.

89
00:03:01.614 --> 00:03:01.794
Yeah.

90
00:03:01.880 --> 00:03:03.661
They go into the low accrual portfolios.

91
00:03:03.716 --> 00:03:04.880
Those are the potential longs.

92
00:03:05.255 --> 00:03:06.427
And the opposite for shorts.

93
00:03:07.098 --> 00:03:10.802
Firms with large positive accruals relative to their absolute earnings,

94
00:03:10.848 --> 00:03:12.895
where reported income is much higher than cash flow,

95
00:03:13.364 --> 00:03:15.473
they land in the high accrual portfolios,

96
00:03:15.474 --> 00:03:16.364
the potential shorts.

97
00:03:16.505 --> 00:03:17.958
Can you give us a feel for the numbers?

98
00:03:17.989 --> 00:03:18.692
Like what kind of

99
00:03:18.728 --> 00:03:22.692
percent accrual value puts a firm in that bottom low accrual group?

100
00:03:23.133 --> 00:03:23.352
Sure.

101
00:03:23.915 --> 00:03:26.133
The paper mentions that for the lowest decile,

102
00:03:26.434 --> 00:03:29.079
the 10th of firms with the lowest percent operating accruals,

103
00:03:29.540 --> 00:03:31.618
the values were typically in the range of,

104
00:03:32.883 --> 00:03:33.344
let's see,

105
00:03:33.985 --> 00:03:37.594
minus 5.7 down to about minus 3.1.

106
00:03:37.719 --> 00:03:39.766
And what's interesting about the firms in that bucket?

107
00:03:39.985 --> 00:03:44.016
What's really key here is that these firms generally had positive cash from operations,

108
00:03:44.610 --> 00:03:46.782
but they also had large negative accruals.

109
00:03:46.992 --> 00:03:48.914
which pull their net income figure down,

110
00:03:49.334 --> 00:03:50.355
sometimes close to zero.

111
00:03:50.717 --> 00:03:53.420
That seems like a crucial distinction from the traditional measure.

112
00:03:53.697 --> 00:03:54.178
Absolutely.

113
00:03:54.381 --> 00:03:57.381
Because with traditional accruals scaled by assets,

114
00:03:57.920 --> 00:04:03.131
that lowest decile could often contain firms with really big losses and significant negative cash flows.

115
00:04:03.803 --> 00:04:06.568
Percent accruals seems to filter those out of the long portfolio.

116
00:04:06.990 --> 00:04:10.115
The paper used Georgia Pacific and supervision as examples,

117
00:04:10.162 --> 00:04:10.490
didn't they,

118
00:04:10.537 --> 00:04:11.334
to highlight this?

119
00:04:11.459 --> 00:04:11.771
They did.

120
00:04:11.990 --> 00:04:12.943
It's a great illustration.

121
00:04:13.459 --> 00:04:14.225
Both firms,

122
00:04:14.506 --> 00:04:16.584
despite being very different in size,

123
00:04:16.980 --> 00:04:20.844
ended up in the lowest percent operating accruals decile in 2001.

124
00:04:21.544 --> 00:04:23.508
Both had strong positive cash flow,

125
00:04:24.008 --> 00:04:25.610
but big negative accruals.

126
00:04:25.946 --> 00:04:29.313
So the percent measure picked up on that similar earnings composition,

127
00:04:29.673 --> 00:04:32.477
regardless of the fact Georgia Pacific was way bigger overall.

128
00:04:32.579 --> 00:04:33.032
Exactly.

129
00:04:33.251 --> 00:04:36.594
Traditional accruals being scaled by assets would treat them very differently.

130
00:04:37.219 --> 00:04:43.001
Percent accruals focuses purely on that cash versus non-cash makeup of the reported profit.

131
00:04:43.484 --> 00:04:43.604
Oh,

132
00:04:43.605 --> 00:04:43.824
okay.

133
00:04:43.845 --> 00:04:45.126
So we have this new measure,

134
00:04:45.166 --> 00:04:46.808
this different way of slicing the data.

135
00:04:47.469 --> 00:04:48.390
Now for the payoff,

136
00:04:48.668 --> 00:04:51.133
did it actually lead to better back-tested results?

137
00:04:51.152 --> 00:04:51.910
Let's talk table four.

138
00:04:52.074 --> 00:04:52.230
Right.

139
00:04:52.254 --> 00:04:54.012
Table four is the direct head-to-head.

140
00:04:54.457 --> 00:04:56.715
Panel A has the percent operating accruals results.

141
00:04:56.754 --> 00:04:58.957
Panel B has the traditional operating accruals.

142
00:04:59.387 --> 00:04:59.996
Same period,

143
00:05:00.402 --> 00:05:01.574
same methodology otherwise.

144
00:05:01.746 --> 00:05:04.465
What's the first thing that hits you looking at panel A versus panel B?

145
00:05:04.871 --> 00:05:07.465
The hedge return is just dramatically different.

146
00:05:08.012 --> 00:05:08.871
For percent accruals,

147
00:05:09.204 --> 00:05:11.205
The hedge portfolio long of bottom decile,

148
00:05:11.324 --> 00:05:12.384
short the top decile,

149
00:05:12.685 --> 00:05:14.785
returned 11.68%

150
00:05:14.847 --> 00:05:15.226
per year.

151
00:05:15.906 --> 00:05:16.566
And crucially,

152
00:05:16.746 --> 00:05:18.566
this was highly statistically significant,

153
00:05:18.910 --> 00:05:21.128
P-value less than 0.001.

154
00:05:21.129 --> 00:05:21.425
Wow,

155
00:05:21.503 --> 00:05:22.246
nearly 12%

156
00:05:22.628 --> 00:05:23.824
and very significant.

157
00:05:24.230 --> 00:05:28.425
What about the long and short legs individually for the percent accrual strategy?

158
00:05:28.722 --> 00:05:29.410
Also strong.

159
00:05:29.753 --> 00:05:30.722
The long portfolio,

160
00:05:30.785 --> 00:05:32.253
the low percent accrual stocks,

161
00:05:32.566 --> 00:05:34.675
returned a positive 5.53%

162
00:05:34.785 --> 00:05:35.191
per year,

163
00:05:35.597 --> 00:05:36.644
also highly significant,

164
00:05:36.675 --> 00:05:36.785
P0.001.

165
00:05:39.284 --> 00:05:40.144
The short portfolio,

166
00:05:40.184 --> 00:05:41.464
the high percent accrual stocks,

167
00:05:41.804 --> 00:05:45.146
had a significant negative return of made at 6.15%,

168
00:05:45.447 --> 00:05:47.286
P.COL 0.009.

169
00:05:47.287 --> 00:05:48.505
So both sides seem to contribute.

170
00:05:48.607 --> 00:05:48.763
Okay.

171
00:05:48.810 --> 00:05:50.388
Now compare that to panel B,

172
00:05:50.529 --> 00:05:51.747
the traditional accruals.

173
00:05:51.786 --> 00:05:52.388
How did that look?

174
00:05:52.505 --> 00:05:52.646
Well,

175
00:05:52.786 --> 00:05:56.161
the hedge return for traditional accruals was 6.51%.

176
00:05:56.162 --> 00:05:56.333
But,

177
00:05:56.646 --> 00:05:57.208
and this is a big,

178
00:05:57.255 --> 00:05:59.130
but it was not statistically significant.

179
00:05:59.443 --> 00:06:01.146
The P value was 0.186.

180
00:06:01.630 --> 00:06:02.693
So based on their tests,

181
00:06:02.694 --> 00:06:04.911
you couldn't be very confident that the 6.5%

182
00:06:05.208 --> 00:06:06.286
wasn't just random chance.

183
00:06:06.349 --> 00:06:06.724
Pretty much.

184
00:06:07.036 --> 00:06:08.538
At conventional significance levels,

185
00:06:08.557 --> 00:06:10.200
you'd say it wasn't reliably different from zero.

186
00:06:10.338 --> 00:06:11.799
And the long short legs for traditional.

187
00:06:12.061 --> 00:06:13.381
This is where it gets really interesting.

188
00:06:13.643 --> 00:06:18.045
The long portfolio for traditional accruals returned only 1.27%

189
00:06:18.084 --> 00:06:18.467
per year,

190
00:06:18.788 --> 00:06:20.045
and that was highly insignificant.

191
00:06:20.209 --> 00:06:21.350
P0.5 years,

192
00:06:21.373 --> 00:06:21.530
zero.

193
00:06:21.913 --> 00:06:22.514
Basically zero.

194
00:06:22.515 --> 00:06:23.108
Wow.

195
00:06:23.186 --> 00:06:23.327
Yeah.

196
00:06:23.873 --> 00:06:24.498
The short side,

197
00:06:24.499 --> 00:06:24.858
however,

198
00:06:25.139 --> 00:06:25.577
did work.

199
00:06:26.030 --> 00:06:30.655
Selling the highest traditional accrual stocks yielded a significant negative 5.24%

200
00:06:30.905 --> 00:06:31.233
per year.

201
00:06:31.764 --> 00:06:34.280
P equals 0.029.

202
00:06:34.928 --> 00:06:37.230
Similar negative return to the percent accruals short side.

203
00:06:37.370 --> 00:06:37.492
OK,

204
00:06:37.591 --> 00:06:38.591
so pulling that together,

205
00:06:38.992 --> 00:06:42.253
it looks like the percent accruals strategy generated a much stronger,

206
00:06:42.335 --> 00:06:44.316
much more reliable hedge return overall.

207
00:06:44.394 --> 00:06:47.800
And the big driver of that difference seems to be the long side,

208
00:06:47.839 --> 00:06:48.042
right?

209
00:06:48.441 --> 00:06:52.589
Buying low percent accruals stocks worked much better than buying low traditional accruals stocks.

210
00:06:52.620 --> 00:06:54.308
It's exactly the story table foretells.

211
00:06:54.714 --> 00:06:56.167
That long portfolio difference,

212
00:06:56.308 --> 00:06:57.574
5.53%

213
00:06:57.745 --> 00:06:59.417
versus 1.27%

214
00:06:59.511 --> 00:07:00.495
is really striking.

215
00:07:00.730 --> 00:07:03.558
It suggests percent accruals is much better at identifying

216
00:07:03.848 --> 00:07:05.229
potentially underpriced stocks.

217
00:07:05.490 --> 00:07:07.573
Did they run the same comparison for total accruals,

218
00:07:07.592 --> 00:07:08.674
not just operating accruals?

219
00:07:08.752 --> 00:07:09.073
They did,

220
00:07:09.135 --> 00:07:09.354
yes.

221
00:07:09.514 --> 00:07:10.334
That's in table five.

222
00:07:10.834 --> 00:07:12.659
And the story is pretty much the same,

223
00:07:12.838 --> 00:07:13.416
quite consistent.

224
00:07:13.495 --> 00:07:13.698
Okay.

225
00:07:13.760 --> 00:07:14.502
What did they find there?

226
00:07:14.799 --> 00:07:16.221
For percent total accruals,

227
00:07:16.362 --> 00:07:18.120
they again found a significant hedge return,

228
00:07:18.121 --> 00:07:19.799
8.53%

229
00:07:19.924 --> 00:07:20.299
per year,

230
00:07:20.784 --> 00:07:21.909
peak of 0.000.

231
00:07:22.346 --> 00:07:22.580
And again,

232
00:07:22.581 --> 00:07:23.580
the long side was strong,

233
00:07:23.862 --> 00:07:24.955
a significant positive,

234
00:07:25.002 --> 00:07:25.909
5.49%,

235
00:07:25.910 --> 00:07:27.346
peak of 0.005.

236
00:07:27.830 --> 00:07:29.534
And traditional total accruals,

237
00:07:29.549 --> 00:07:29.987
let me guess,

238
00:07:30.596 --> 00:07:30.877
weaker.

239
00:07:30.987 --> 00:07:31.455
You got it.

240
00:07:31.764 --> 00:07:34.388
The hedge return for traditional total accruals was insignificant.

241
00:07:34.888 --> 00:07:37.149
The long portfolio return was also insignificant,

242
00:07:37.169 --> 00:07:37.849
just 1.60%.

243
00:07:38.911 --> 00:07:40.610
The short side did show significance again,

244
00:07:40.735 --> 00:07:42.110
negative 4.23%,

245
00:07:42.313 --> 00:07:43.813
P yield 0.048.

246
00:07:44.313 --> 00:07:46.821
So it really seems robust across both definitions,

247
00:07:46.923 --> 00:07:48.016
operating and total.

248
00:07:48.345 --> 00:07:50.657
This percent scaling seems to unlock more performance,

249
00:07:50.720 --> 00:07:52.329
particularly from the long book.

250
00:07:52.563 --> 00:07:54.673
That's certainly what these back tests strongly indicate.

251
00:07:54.813 --> 00:07:58.423
Did the paper look at what kinds of companies end up in these extreme deciles?

252
00:07:58.454 --> 00:08:00.095
Are they different based on the measure used?

253
00:08:00.304 --> 00:08:00.444
Yes.

254
00:08:00.504 --> 00:08:01.926
Table 2 provides some characteristics,

255
00:08:02.006 --> 00:08:03.186
and there's some interesting differences,

256
00:08:03.247 --> 00:08:04.470
especially in that lowest decile,

257
00:08:04.471 --> 00:08:05.489
the potential longs.

258
00:08:05.891 --> 00:08:06.212
That's what?

259
00:08:06.731 --> 00:08:06.872
Well,

260
00:08:06.891 --> 00:08:07.192
for one,

261
00:08:07.212 --> 00:08:10.891
the average market cap of firms in the lowest percent accrual decile was much larger,

262
00:08:10.938 --> 00:08:12.095
about $1.5 billion,

263
00:08:12.915 --> 00:08:17.579
compared to only $474 million for the lowest traditional accrual decile.

264
00:08:17.954 --> 00:08:18.157
Ah,

265
00:08:18.829 --> 00:08:20.345
that's potentially quite practical,

266
00:08:20.407 --> 00:08:20.704
isn't it?

267
00:08:20.845 --> 00:08:22.720
Larger firms usually mean better liquidity,

268
00:08:22.970 --> 00:08:25.610
maybe a lower trading cost to implement the strategy.

269
00:08:25.782 --> 00:08:26.251
Exactly.

270
00:08:26.532 --> 00:08:27.688
That could be a real advantage.

271
00:08:28.168 --> 00:08:28.388
Also,

272
00:08:28.488 --> 00:08:34.055
those low percent accrual firms tended to have higher average return on assets and,

273
00:08:34.555 --> 00:08:35.434
maybe more importantly,

274
00:08:35.817 --> 00:08:37.255
higher cash return on assets.

275
00:08:37.536 --> 00:08:38.778
Suggesting they are actually,

276
00:08:39.239 --> 00:08:39.380
well,

277
00:08:39.622 --> 00:08:43.083
fundamentally healthier businesses on average in that long portfolio.

278
00:08:43.145 --> 00:08:44.122
It seems that way.

279
00:08:44.606 --> 00:08:49.575
It suggests percent accruals might be better at picking firms that are performing okay on a cash basis,

280
00:08:49.934 --> 00:08:53.028
but where the accounting is perhaps temporarily obscuring that.

281
00:08:53.200 --> 00:08:55.653
That ties back nicely to the original rationale.

282
00:08:55.762 --> 00:08:55.872
And

283
00:08:56.168 --> 00:08:58.711
Did they check how much overlap there actually is between the two measures?

284
00:08:58.731 --> 00:09:00.053
Are they picking the same stocks?

285
00:09:00.354 --> 00:09:00.912
Good question.

286
00:09:01.373 --> 00:09:02.955
Table 3 looks at that overlap.

287
00:09:03.475 --> 00:09:04.799
And it's surprisingly low,

288
00:09:05.018 --> 00:09:07.197
especially in those crucial extreme deciles.

289
00:09:07.658 --> 00:09:08.104
For instance,

290
00:09:08.322 --> 00:09:10.424
only about 11.78%

291
00:09:10.425 --> 00:09:16.244
of the firms in the lowest decile using percent accruals were also in the lowest decile using traditional accruals.

292
00:09:16.463 --> 00:09:16.729
Wow.

293
00:09:16.963 --> 00:09:17.807
Only about 12%

294
00:09:18.135 --> 00:09:18.525
overlap.

295
00:09:18.775 --> 00:09:21.307
That really drives home the point that this isn't just a minor tweak.

296
00:09:21.432 --> 00:09:23.713
It's identifying a substantially different group of stocks.

297
00:09:24.072 --> 00:09:24.541
Precisely.

298
00:09:25.316 --> 00:09:27.559
It's finding a different set of potential opportunities.

299
00:09:27.758 --> 00:09:29.520
And like any good academic paper,

300
00:09:29.602 --> 00:09:34.067
they surely ran some robustness checks to see if the results held up under different conditions.

301
00:09:34.207 --> 00:09:34.324
Oh,

302
00:09:34.387 --> 00:09:34.528
yes,

303
00:09:34.543 --> 00:09:35.207
they did several.

304
00:09:35.449 --> 00:09:35.926
For example,

305
00:09:35.965 --> 00:09:41.988
Table 10 examined whether including or excluding special items from the earnings calculation changed things.

306
00:09:42.660 --> 00:09:45.551
They found the percent accrual strategy worked well either way.

307
00:09:45.879 --> 00:09:46.535
Which is good.

308
00:09:46.551 --> 00:09:50.582
It means it's not just driven by weird one-off accounting things.

309
00:09:50.723 --> 00:09:50.957
Right.

310
00:09:51.098 --> 00:09:53.317
And Table 11 looked at performance separately.

311
00:09:53.652 --> 00:09:56.354
for firms making profits versus firms making losses.

312
00:09:57.016 --> 00:09:59.338
Percent accrual showed effectiveness for both groups.

313
00:09:59.897 --> 00:10:04.506
That's interesting because the traditional anomaly has sometimes been shown to be weaker among loss-making firms.

314
00:10:04.881 --> 00:10:05.006
OK,

315
00:10:05.225 --> 00:10:06.326
another point in its favor.

316
00:10:06.467 --> 00:10:07.022
Anything else?

317
00:10:07.287 --> 00:10:10.592
They also looked at performance across different levels of arbitrage risk,

318
00:10:10.654 --> 00:10:11.014
basically,

319
00:10:11.092 --> 00:10:13.858
how easy or hard it is to trade these stocks.

320
00:10:14.775 --> 00:10:18.919
The percent accrual strategy seemed to remain significant across most of these risk levels,

321
00:10:19.321 --> 00:10:21.243
suggesting it's not just working in tiny,

322
00:10:21.583 --> 00:10:22.083
illiquid,

323
00:10:22.321 --> 00:10:23.344
hard to trade stocks.

324
00:10:23.680 --> 00:10:24.266
That's important,

325
00:10:24.305 --> 00:10:24.501
too.

326
00:10:24.727 --> 00:10:26.547
It suggests broader applicability.

327
00:10:27.126 --> 00:10:27.251
OK,

328
00:10:27.407 --> 00:10:30.469
so wrapping this all up after diving into this paper,

329
00:10:30.907 --> 00:10:33.907
what's the main message for an algo trader listening today?

330
00:10:34.329 --> 00:10:37.141
I think the key takeaway is that the evidence here,

331
00:10:37.438 --> 00:10:38.516
based on their back tests,

332
00:10:38.532 --> 00:10:39.485
it's pretty compelling.

333
00:10:40.126 --> 00:10:41.641
Using percent accruals as a signal

334
00:10:42.143 --> 00:10:44.747
particularly for identifying potential long candidates,

335
00:10:45.407 --> 00:10:50.051
appears to offer a potentially more profitable approach compared to the traditional accruals measure.

336
00:10:50.454 --> 00:10:51.352
The performance,

337
00:10:51.415 --> 00:10:52.735
especially on that long side,

338
00:10:52.750 --> 00:10:54.735
was significantly better in their tests.

339
00:10:54.899 --> 00:10:55.133
Yes,

340
00:10:55.555 --> 00:10:58.055
significantly better and more statistically robust.

341
00:10:58.782 --> 00:11:05.188
It suggests this alternative definition might capture something about earnings quality or investor misperception more effectively.

342
00:11:05.547 --> 00:11:10.391
So for traders looking to refine existing quality or value strategies or develop new ones,

343
00:11:10.803 --> 00:11:13.406
This percent accruals idea definitely warrants a closer look.

344
00:11:13.707 --> 00:11:14.146
Absolutely.

345
00:11:14.508 --> 00:11:23.094
The paper hints that its success might come from better identifying situations where investors underestimate the persistence of cash flows versus accruals.

346
00:11:23.515 --> 00:11:26.476
And maybe the fact that it relies less on very small,

347
00:11:26.906 --> 00:11:29.015
potentially high-risk stocks is also a factor.

348
00:11:29.281 --> 00:11:31.328
Definitely food for thought and further testing.

349
00:11:31.687 --> 00:11:34.031
Thank you for tuning in to Papers with Backtest podcast.

350
00:11:34.265 --> 00:11:36.453
We hope today's episode gave you useful insights.

351
00:11:36.781 --> 00:11:38.672
Join us next time as we break down more research.

352
00:11:38.915 --> 00:11:40.356
And for more papers and backtests,

353
00:11:40.497 --> 00:11:43.280
find us at https.paperswithbacktests.com.

354
00:11:43.562 --> 00:11:44.101
Happy trading.