1
00:00:00,500 --> 00:00:05,000
The global economic financial crisis has reignited public interest

2
00:00:05,000 --> 00:00:08,416
in something that's actually one of the oldest questions in economics,

3
00:00:08,416 --> 00:00:11,100
dating back to at least before Adam Smith.

4
00:00:11,100 --> 00:00:16,584
And that is, why is it that countries with seemingly similar economies and institutions

5
00:00:16,584 --> 00:00:19,833
can display radically different savings behavior?

6
00:00:19,833 --> 00:00:24,367
Now, many brilliant economists have spent their entire lives working on this question,

7
00:00:24,367 --> 00:00:27,784
and as a field we've made a tremendous amount of headway

8
00:00:27,784 --> 00:00:30,217
and we understand a lot about this.

9
00:00:30,217 --> 00:00:33,851
What I'm here to talk with you about today is an intriguing new hypothesis

10
00:00:33,851 --> 00:00:37,883
and some surprisingly powerful new findings that I've been working on

11
00:00:37,883 --> 00:00:42,589
about the link between the structure of the language you speak

12
00:00:42,604 --> 00:00:47,000
and how you find yourself with the propensity to save.

13
00:00:47,000 --> 00:00:50,067
Let me tell you a little bit about savings rates, a little bit about language,

14
00:00:50,067 --> 00:00:52,417
and then I'll draw that connection.

15
00:00:52,417 --> 00:00:56,984
Let's start by thinking about the member countries of the OECD,

16
00:00:56,984 --> 00:01:00,285
or the Organization of Economic Cooperation and Development.

17
00:01:00,285 --> 00:01:04,184
OECD countries, by and large, you should think about these

18
00:01:04,184 --> 00:01:06,822
as the richest, most industrialized countries in the world.

19
00:01:06,822 --> 00:01:10,872
And by joining the OECD, they were affirming a common commitment

20
00:01:10,872 --> 00:01:14,310
to democracy, open markets and free trade.

21
00:01:14,310 --> 00:01:18,995
Despite all of these similarities, we see huge differences in savings behavior.

22
00:01:18,995 --> 00:01:21,445
So all the way over on the left of this graph,

23
00:01:21,445 --> 00:01:26,179
what you see is many OECD countries saving over a quarter of their GDP every year,

24
00:01:26,179 --> 00:01:30,860
and some OECD countries saving over a third of their GDP per year.

25
00:01:30,860 --> 00:01:35,628
Holding down the right flank of the OECD, all the way on the other side, is Greece.

26
00:01:35,628 --> 00:01:39,044
And what you can see is that over the last 25 years,

27
00:01:39,044 --> 00:01:42,944
Greece has barely managed to save more than 10 percent of their GDP.

28
00:01:42,944 --> 00:01:49,818
It should be noted, of course, that the United States and the U.K. are the next in line.

29
00:01:49,818 --> 00:01:52,396
Now that we see these huge differences in savings rates,

30
00:01:52,396 --> 00:01:56,062
how is it possible that language might have something to do with these differences?

31
00:01:56,062 --> 00:01:59,111
Let me tell you a little bit about how languages fundamentally differ.

32
00:01:59,111 --> 00:02:04,678
Linguists and cognitive scientists have been exploring this question for many years now.

33
00:02:04,678 --> 00:02:09,288
And then I'll draw the connection between these two behaviors.

34
00:02:09,288 --> 00:02:11,896
Many of you have probably already noticed that I'm Chinese.

35
00:02:11,896 --> 00:02:14,861
I grew up in the Midwest of the United States.

36
00:02:14,861 --> 00:02:17,312
And something I realized quite early on

37
00:02:17,312 --> 00:02:20,903
was that the Chinese language forced me to speak about and --

38
00:02:20,903 --> 00:02:23,794
in fact, more fundamentally than that --

39
00:02:23,794 --> 00:02:27,884
ever so slightly forced me to think about family in very different ways.

40
00:02:27,884 --> 00:02:29,961
Now, how might that be? Let me give you an example.

41
00:02:29,961 --> 00:02:34,363
Suppose I were talking with you and I was introducing you to my uncle.

42
00:02:34,363 --> 00:02:37,229
You understood exactly what I just said in English.

43
00:02:37,229 --> 00:02:40,179
If we were speaking Mandarin Chinese with each other, though,

44
00:02:40,179 --> 00:02:42,245
I wouldn't have that luxury.

45
00:02:42,245 --> 00:02:45,078
I wouldn't have been able to convey so little information.

46
00:02:45,078 --> 00:02:47,462
What my language would have forced me to do,

47
00:02:47,462 --> 00:02:49,462
instead of just telling you, "This is my uncle,"

48
00:02:49,462 --> 00:02:52,744
is to tell you a tremendous amount of additional information.

49
00:02:52,744 --> 00:02:54,563
My language would force me to tell you

50
00:02:54,563 --> 00:02:57,979
whether or not this was an uncle on my mother's side or my father's side,

51
00:02:57,979 --> 00:03:01,063
whether this was an uncle by marriage or by birth,

52
00:03:01,063 --> 00:03:03,295
and if this man was my father's brother,

53
00:03:03,295 --> 00:03:06,079
whether he was older than or younger than my father.

54
00:03:06,079 --> 00:03:10,212
All of this information is obligatory. Chinese doesn't let me ignore it.

55
00:03:10,212 --> 00:03:12,378
And in fact, if I want to speak correctly,

56
00:03:12,378 --> 00:03:15,495
Chinese forces me to constantly think about it.

57
00:03:15,495 --> 00:03:19,444
Now, that fascinated me endlessly as a child,

58
00:03:19,444 --> 00:03:22,679
but what fascinates me even more today as an economist

59
00:03:22,679 --> 00:03:27,980
is that some of these same differences carry through to how languages speak about time.

60
00:03:27,980 --> 00:03:32,229
So for example, if I'm speaking in English, I have to speak grammatically differently

61
00:03:32,229 --> 00:03:34,962
if I'm talking about past rain, "It rained yesterday,"

62
00:03:34,962 --> 00:03:37,194
current rain, "It is raining now,"

63
00:03:37,194 --> 00:03:39,628
or future rain, "It will rain tomorrow."

64
00:03:39,628 --> 00:03:44,455
Notice that English requires a lot more information with respect to the timing of events.

65
00:03:44,455 --> 00:03:46,462
Why? Because I have to consider that

66
00:03:46,462 --> 00:03:51,245
and I have to modify what I'm saying to say, "It will rain," or "It's going to rain."

67
00:03:51,245 --> 00:03:55,361
It's simply not permissible in English to say, "It rain tomorrow."

68
00:03:55,361 --> 00:03:59,545
In contrast to that, that's almost exactly what you would say in Chinese.

69
00:03:59,545 --> 00:04:01,861
A Chinese speaker can basically say something

70
00:04:01,861 --> 00:04:04,445
that sounds very strange to an English speaker's ears.

71
00:04:04,445 --> 00:04:09,012
They can say, "Yesterday it rain," "Now it rain," "Tomorrow it rain."

72
00:04:09,012 --> 00:04:12,875
In some deep sense, Chinese doesn't divide up the time spectrum

73
00:04:12,875 --> 00:04:19,308
in the same way that English forces us to constantly do in order to speak correctly.

74
00:04:19,308 --> 00:04:20,892
Is this difference in languages

75
00:04:20,892 --> 00:04:25,087
only between very, very distantly related languages, like English and Chinese?

76
00:04:25,087 --> 00:04:26,045
Actually, no.

77
00:04:26,045 --> 00:04:29,778
So many of you know, in this room, that English is a Germanic language.

78
00:04:29,778 --> 00:04:33,693
What you may not have realized is that English is actually an outlier.

79
00:04:33,693 --> 00:04:36,943
It is the only Germanic language that requires this.

80
00:04:36,943 --> 00:04:39,827
For example, most other Germanic language speakers

81
00:04:39,827 --> 00:04:42,844
feel completely comfortable talking about rain tomorrow

82
00:04:42,844 --> 00:04:44,810
by saying, "Morgen regnet es,"

83
00:04:44,810 --> 00:04:48,710
quite literally to an English ear, "It rain tomorrow."

84
00:04:48,710 --> 00:04:53,777
This led me, as a behavioral economist, to an intriguing hypothesis.

85
00:04:53,777 --> 00:04:58,026
Could how you speak about time, could how your language forces you to think about time,

86
00:04:58,026 --> 00:05:01,777
affect your propensity to behave across time?

87
00:05:01,777 --> 00:05:04,677
You speak English, a futured language.

88
00:05:04,677 --> 00:05:07,844
And what that means is that every time you discuss the future,

89
00:05:07,844 --> 00:05:09,411
or any kind of a future event,

90
00:05:09,411 --> 00:05:12,810
grammatically you're forced to cleave that from the present

91
00:05:12,810 --> 00:05:15,443
and treat it as if it's something viscerally different.

92
00:05:15,443 --> 00:05:17,943
Now suppose that that visceral difference

93
00:05:17,943 --> 00:05:22,143
makes you subtly dissociate the future from the present every time you speak.

94
00:05:22,143 --> 00:05:24,056
If that's true and it makes the future feel

95
00:05:24,056 --> 00:05:27,039
like something more distant and more different from the present,

96
00:05:27,039 --> 00:05:29,693
that's going to make it harder to save.

97
00:05:29,693 --> 00:05:32,244
If, on the other hand, you speak a futureless language,

98
00:05:32,244 --> 00:05:35,627
the present and the future, you speak about them identically.

99
00:05:35,627 --> 00:05:38,610
If that subtly nudges you to feel about them identically,

100
00:05:38,610 --> 00:05:40,994
that's going to make it easier to save.

101
00:05:40,994 --> 00:05:43,544
Now this is a fanciful theory.

102
00:05:43,544 --> 00:05:46,444
I'm a professor, I get paid to have fanciful theories.

103
00:05:46,444 --> 00:05:50,678
But how would you actually go about testing such a theory?

104
00:05:50,678 --> 00:05:54,744
Well, what I did with that was to access the linguistics literature.

105
00:05:54,744 --> 00:05:59,127
And interestingly enough, there are pockets of futureless language speakers

106
00:05:59,127 --> 00:06:01,060
situated all over the world.

107
00:06:01,060 --> 00:06:04,426
This is a pocket of futureless language speakers in Northern Europe.

108
00:06:04,426 --> 00:06:07,327
Interestingly enough, when you start to crank the data,

109
00:06:07,327 --> 00:06:10,560
these pockets of futureless language speakers all around the world

110
00:06:10,560 --> 00:06:14,494
turn out to be, by and large, some of the world's best savers.

111
00:06:14,494 --> 00:06:16,660
Just to give you a hint of that,

112
00:06:16,660 --> 00:06:19,410
let's look back at that OECD graph that we were talking about.

113
00:06:19,410 --> 00:06:22,794
What you see is that these bars are systematically taller

114
00:06:22,794 --> 00:06:24,926
and systematically shifted to the left

115
00:06:24,926 --> 00:06:29,444
compared to these bars which are the members of the OECD that speak futured languages.

116
00:06:29,444 --> 00:06:30,907
What is the average difference here?

117
00:06:30,907 --> 00:06:34,193
Five percentage points of your GDP saved per year.

118
00:06:34,193 --> 00:06:38,927
Over 25 years that has huge long-run effects on the wealth of your nation.

119
00:06:38,927 --> 00:06:41,627
Now while these findings are suggestive,

120
00:06:41,627 --> 00:06:43,693
countries can be different in so many different ways

121
00:06:43,693 --> 00:06:48,077
that it's very, very difficult sometimes to account for all of these possible differences.

122
00:06:48,077 --> 00:06:52,109
What I'm going to show you, though, is something that I've been engaging in for a year,

123
00:06:52,109 --> 00:06:54,432
which is trying to gather all of the largest datasets

124
00:06:54,432 --> 00:06:56,724
that we have access to as economists,

125
00:06:56,724 --> 00:07:00,106
and I'm going to try and strip away all of those possible differences,

126
00:07:00,106 --> 00:07:02,760
hoping to get this relationship to break.

127
00:07:02,760 --> 00:07:07,791
And just in summary, no matter how far I push this, I can't get it to break.

128
00:07:07,791 --> 00:07:09,556
Let me show you how far you can do that.

129
00:07:09,556 --> 00:07:14,189
One way to imagine that is I gather large datasets from around the world.

130
00:07:14,189 --> 00:07:17,923
So for example, there is the Survey of Health, [Aging] and Retirement in Europe.

131
00:07:17,923 --> 00:07:21,757
From this dataset you actually learn that retired European families

132
00:07:21,757 --> 00:07:24,390
are extremely patient with survey takers.

133
00:07:24,390 --> 00:07:26,306
(Laughter)

134
00:07:26,306 --> 00:07:30,690
So imagine that you're a retired household in Belgium and someone comes to your front door.

135
00:07:30,690 --> 00:07:35,274
"Excuse me, would you mind if I peruse your stock portfolio?

136
00:07:35,274 --> 00:07:38,806
Do you happen to know how much your house is worth? Do you mind telling me?

137
00:07:38,806 --> 00:07:42,073
Would you happen to have a hallway that's more than 10 meters long?

138
00:07:42,073 --> 00:07:46,574
If you do, would you mind if I timed how long it took you to walk down that hallway?

139
00:07:46,574 --> 00:07:50,471
Would you mind squeezing as hard as you can, in your dominant hand, this device

140
00:07:50,471 --> 00:07:51,983
so I can measure your grip strength?

141
00:07:51,983 --> 00:07:56,006
How about blowing into this tube so I can measure your lung capacity?"

142
00:07:56,006 --> 00:07:58,890
The survey takes over a day.

143
00:07:58,890 --> 00:08:00,373
(Laughter)

144
00:08:00,373 --> 00:08:04,273
Combine that with a Demographic and Health Survey

145
00:08:04,273 --> 00:08:08,723
collected by USAID in developing countries in Africa, for example,

146
00:08:08,723 --> 00:08:13,874
which that survey actually can go so far as to directly measure the HIV status

147
00:08:13,874 --> 00:08:16,973
of families living in, for example, rural Nigeria.

148
00:08:16,973 --> 00:08:18,874
Combine that with a world value survey,

149
00:08:18,874 --> 00:08:23,307
which measures the political opinions and, fortunately for me, the savings behaviors

150
00:08:23,307 --> 00:08:28,006
of millions of families in hundreds of countries around the world.

151
00:08:28,006 --> 00:08:31,824
Take all of that data, combine it, and this map is what you get.

152
00:08:31,824 --> 00:08:34,074
What you find is nine countries around the world

153
00:08:34,074 --> 00:08:36,726
that have significant native populations

154
00:08:36,726 --> 00:08:40,773
which speak both futureless and futured languages.

155
00:08:40,773 --> 00:08:44,307
And what I'm going to do is form statistical matched pairs

156
00:08:44,307 --> 00:08:49,900
between families that are nearly identical on every dimension that I can measure,

157
00:08:49,900 --> 00:08:53,437
and then I'm going to explore whether or not the link between language and savings holds

158
00:08:53,437 --> 00:08:56,920
even after controlling for all of these levels.

159
00:08:56,920 --> 00:08:59,137
What are the characteristics we can control for?

160
00:08:59,137 --> 00:09:01,901
Well I'm going to match families on country of birth and residence,

161
00:09:01,901 --> 00:09:04,303
the demographics -- what sex, their age --

162
00:09:04,303 --> 00:09:06,437
their income level within their own country,

163
00:09:06,437 --> 00:09:09,452
their educational achievement, a lot about their family structure.

164
00:09:09,452 --> 00:09:12,970
It turns out there are six different ways to be married in Europe.

165
00:09:12,970 --> 00:09:17,170
And most granularly, I break them down by religion

166
00:09:17,170 --> 00:09:20,485
where there are 72 categories of religions in the world --

167
00:09:20,485 --> 00:09:22,202
so an extreme level of granularity.

168
00:09:22,202 --> 00:09:26,735
There are 1.4 billion different ways that a family can find itself.

169
00:09:26,735 --> 00:09:30,784
Now effectively everything I'm going to tell you from now on

170
00:09:30,784 --> 00:09:33,834
is only comparing these basically nearly identical families.

171
00:09:33,834 --> 00:09:36,334
It's getting as close as possible to the thought experiment

172
00:09:36,334 --> 00:09:39,267
of finding two families both of whom live in Brussels

173
00:09:39,267 --> 00:09:42,267
who are identical on every single one of these dimensions,

174
00:09:42,267 --> 00:09:45,383
but one of whom speaks Flemish and one of whom speaks French;

175
00:09:45,383 --> 00:09:48,100
or two families that live in a rural district in Nigeria,

176
00:09:48,100 --> 00:09:51,933
one of whom speaks Hausa and one of whom speaks Igbo.

177
00:09:51,933 --> 00:09:55,817
Now even after all of this granular level of control,

178
00:09:55,817 --> 00:09:58,983
do futureless language speakers seem to save more?

179
00:09:58,983 --> 00:10:02,636
Yes, futureless language speakers, even after this level of control,

180
00:10:02,636 --> 00:10:06,334
are 30 percent more likely to report having saved in any given year.

181
00:10:06,334 --> 00:10:08,150
Does this have cumulative effects?

182
00:10:08,150 --> 00:10:12,570
Yes, by the time they retire, futureless language speakers, holding constant their income,

183
00:10:12,570 --> 00:10:15,638
are going to retire with 25 percent more in savings.

184
00:10:15,638 --> 00:10:18,119
Can we push this data even further?

185
00:10:18,119 --> 00:10:23,418
Yes, because I just told you, we actually collect a lot of health data as economists.

186
00:10:23,418 --> 00:10:27,301
Now how can we think about health behaviors to think about savings?

187
00:10:27,301 --> 00:10:30,134
Well, think about smoking, for example.

188
00:10:30,134 --> 00:10:33,317
Smoking is in some deep sense negative savings.

189
00:10:33,317 --> 00:10:36,983
If savings is current pain in exchange for future pleasure,

190
00:10:36,983 --> 00:10:38,291
smoking is just the opposite.

191
00:10:38,291 --> 00:10:41,150
It's current pleasure in exchange for future pain.

192
00:10:41,150 --> 00:10:44,100
What we should expect then is the opposite effect.

193
00:10:44,100 --> 00:10:45,868
And that's exactly what we find.

194
00:10:45,868 --> 00:10:49,635
Futureless language speakers are 20 to 24 percent less likely

195
00:10:49,635 --> 00:10:53,050
to be smoking at any given point in time compared to identical families,

196
00:10:53,050 --> 00:10:55,951
and they're going to be 13 to 17 percent less likely

197
00:10:55,951 --> 00:10:58,168
to be obese by the time they retire,

198
00:10:58,168 --> 00:11:00,631
and they're going to report being 21 percent more likely

199
00:11:00,631 --> 00:11:02,918
to have used a condom in their last sexual encounter.

200
00:11:02,918 --> 00:11:06,401
I could go on and on with the list of differences that you can find.

201
00:11:06,401 --> 00:11:10,201
It's almost impossible not to find a savings behavior

202
00:11:10,201 --> 00:11:12,800
for which this strong effect isn't present.

203
00:11:12,800 --> 00:11:17,550
My linguistics and economics colleagues at Yale and I are just starting to do this work

204
00:11:17,550 --> 00:11:22,717
and really explore and understand the ways that these subtle nudges

205
00:11:22,717 --> 00:11:28,112
cause us to think more or less about the future every single time we speak.

206
00:11:28,112 --> 00:11:30,413
Ultimately, the goal,

207
00:11:30,413 --> 00:11:34,612
once we understand how these subtle effects can change our decision making,

208
00:11:34,612 --> 00:11:37,562
we want to be able to provide people tools

209
00:11:37,562 --> 00:11:40,370
so that they can consciously make themselves better savers

210
00:11:40,370 --> 00:11:43,629
and more conscious investors in their own future.

211
00:11:43,629 --> 00:11:45,896
Thank you very much.

212
00:11:45,896 --> 00:11:52,264
(Applause)