Month: November 2013

Nowcasting with Google, by Mahalia Jackman

Nowcasting has become widely popular in economics. In its most basic form, nowcasting can be summarised as ‘predicting the present and sometimes recent past’. At first this may seem a bit strange. Why would economists want to forecast the present, or even the past? Can you even forecast the past?  Economists ‘nowcast’ because of the significant time lag in statistical releases. For instance, inflation statistics for July 2013 were released in October 2013; and don’t even get me started on the national accounts data. We often want to know what is going on in the economy in real time. For example, we want to know what happened in the tourism industry in May on June 1st. So economists began looking for high frequency indicators which tend to be highly correlated with the ‘late release’ data. The problem is, even the leading indicators have a lag…until Google!

Putting Google Data to Work

Recently, Google began to release real-time information on its users’ search queries though its Google Trends interface. The data is of a weekly frequency and dates back to 2004. You can filter the results for the search term by country (i.e. the location where the search was generated), time period and category (for instance, searches in the Google Travel category or Google Finance). This has proven to be a useful data source as it allows researchers to observe interest in a product, brand or society in general. In fact, in a recent paper by Google economists Hyunyoung Choi and Hal Varian titled “Predicting the present with Google trends” illustrated the usefulness of Google data for nowcasting a host of US economic time series, including automobile sales, unemployment claims and even trends in travel.  Since then, several other researchers have explored the use of Google search data as an economic indicator. Google trends has been used to nowcast growth cycles in Israel, private consumption in the US, some indicators of labour and housing markets in the UK, unemployment in Turkey, retail and unemployment in Belgium, automobiles in Chile, and the list goes on.

Applying Google Econometrics to the Caribbean

After reading several studies on the larger economies, I started to think about the applicability in Caribbean.  So I – with the aid of Simon Naitram – put Google econometrics to the test. We focused on whether or not observing internet habits in Barbados main tourism markets can provide insights on trends in tourist arrivals to Barbados[i].

Why tourism? First, Barbados’ economic fortunes are closely tied to its tourism industry and as recent economic activity highlights, if tourism sneezes, the entire country catches pneumonia! Second, tourism is often described as an information intensive industry. Quite a few studies have shown that the internet (particularly search engines) has become one of the most effective means for tourists to seek information on destinations or discounts on hotels and flights. We assumed that Google search queries may be capable of providing information on the level of interest in Barbados, and so serve as a leading indicator of tourism demand.

The Results

So…how predictable is tourism using this algorithm?  Here is one of the examples from the paper. The blue line shows weekly Canadian arrivals and the red dotted line represents Canadian search volume histories related to the simple search term “Barbados” under the Travel category of Google Trends. Even through a simple visual examination there does seem to be a high degree of correlation, and the Google trends series even mirrors the seasonality in Canadian arrivals.

Canada - Mahalia upload

However, just because they are correlated does not mean that Google can ‘nowcast’ tourism. Forecasting 101 tells us that performance is relative: the predictive power of a model is only meaningful in relation to some baseline means of prediction.   So, we proceeded to nowcast Canadian arrivals based on its past values (this is a simplification of the process- to be specific, we used an autoregressive support vector regression that you can read more about in our paper). We then compared them with a model incorporating the Google data (which we deemed the Google-based support vector regression). In a nutshell, if the model with Google provides better predictions, then we can conclude that Google search query volumes add insight on tourism behaviour.

The figure below plots the two nowcasts as well as the actual data over a 12-week period. The red line shows the nowcast of arrivals from Canada with Google, the blue line shows the nowcasts without Google and the grey bars represent the actual data.

nowcasts - Mahalia upload

A key finding from the paper was that the accuracy of Google predictability varies. For some tourist markets (like Canada) Google trends worked very well, while for others (like the US), Google data did not provide any additional information from what can be “learned” from past behaviour.

Google Data Pros and Cons

It should be noted that Google data is a mixed blessing. On the positive side, Google trends data have a large number of characteristics that are appealing; for instance, they contain information on a large portion of internet users, the information is free and it is practically available in real time.  On the other hand, it has a number of shortcomings:

  • Google trends uses relative data rather than absolute. That is, it doesn’t give the total number of searches, but an index which is relative to all search queries in a particular region.
  • Different users interested in the same topic could enter entirely different search queries & different users with entirely different intentions could enter very similar search queries  (though this is not likely to impact what Simon and I did, as we limited the results to only searches in the ‘Travel’ category)
  • Internet searches are correlated with factors such as age, income, etc., which may limit predictability in certain market types
  • Currently, Google can only generate search queries originating from four Caribbean countries – the Dominican Republic, Haiti, Jamaica and Trinidad.

What can we take from this?

  1. Google Econometrics has the potential to act as a leading indicator
  2. Predictability depends – so be sure to test the nowcasts against some benchmark model
  3. Beware of the short-comings and use wisely!


Mahalia Jackman is Head of Model Development at Antilles Economics. She can be contacted at or via her linkedin page.

[i] Jackman, M. and Naitram, S. “Nowcasting tourist arrivals to Barbados – Just Google it!” Paper presented at the Central Bank of Barbados Annual Review Seminar, July 2013.

Barbados Economic Recovery: Framing the Discussion

Last week on a popular radio show in Barbados, two leading economists and a former head of a large company in Barbados discussed and took calls from the public on the Barbados economic recovery. I found the discussion extremely interesting and I commend all of the panelists for their insights. At the same time, however, I found myself wondering if the discussion was framed correctly.

Be careful what questions you ask, you may just get answers

I believe that you only get answers to the questions you ask. So, for example, if I asked you if you thought that an umbrella would help reduce the heat from the sun, we could have a long discussion on the usefulness of umbrellas, their design, the materials they are made out of, when they are most effective, etc. We may not discuss if the temperature was actually high enough that day to warrant a solution in the first place. We may not discuss if there were other solutions to reducing the impact of the sun’s heat. We may not discuss the benefits of exposing oneself to the sun’s rays. We would focus our discussion on answering the question asked: would an umbrella work.

So, when we ask panelists whether Barbados is in a crisis, they respond with their opinions on what defines a crisis and therefore whether they believe we are in one. When we ask them if the government or the private sector should be the one to get us out, they debate the effectiveness of each option as a means to solving the current problems. When we ask them for specific examples of what we should do to get out of the current crisis, they list suggestions from strengthening tourism, cutting government expenditure, and retraining the labour force.

These are all valid points. Very few people would have any major disagreements. Furthermore, all of their suggestions could work. Unfortunately, this leaves us with no clear idea of what to do next and what I consider to be the ‘too much choice’ dilemma: when faced with too many options we opt to choose none.

We need to frame our discussion around solving our long-term or structural problems first before we can identify the best short-term actions. Policy decisions, probably more than most other types of decision-making, is about long-term planning because timing is the most critical factor in macroeconomic policy. The ‘right’ policy implemented at the wrong time will not work just like the ‘wrong’ policy implemented at the right time will not work. The only combination that works is the ‘right’ policy at the right time.

We created our structural problems by answering the wrong questions before. When government needed more revenue, it added the Value Added Tax (VAT), which did not address the underlying problem of insufficient production in the economy. When the VAT was raised, it did not address the underlying problem of declining production and consumption. When unions demanded pay increases, it did not address the challenge of falling corporate revenue and labour productivity. When companies cut marketing budgets and raised prices, they ignored the fact that consumer demand was falling. All of these solutions were aimed at fixing some short-term challenge. But in the end, they deepened structural weaknesses and contributed to yet another round of problems.

Solving our structural challenges requires us to ask the right questions. What if instead we had asked the panelists for their view of the ‘ideal’ economy of Barbados before we asked any other question? Maybe they would say that it was less dependent on tourism; the government was smaller; the labour force was more flexible; the exports were more diversified; the infrastructure and institutions were stronger; or some other goal or combination of goals. Whatever the goal, we can then debate how we achieve it.

All other questions, and their answers, should be framed with our long-term goals in mind. Suppose we determine that a smaller government was critical to our vision of the ‘ideal’ Barbados economy. Maybe we would be much more supportive of cutting the size of the civil service, embracing technology for standardised government services and divesting of government-owned companies.

The same thinking applies to becoming less dependent on tourism, for example. With this as our goal, maybe we would be hesitant to grant additional subsidies and more financial resources to the hoteliers. Maybe we would revisit our legislation for our international financial services sector to encourage growth in this industry. Maybe we would embrace science and technology more in our agricultural sector to compensate for the physical challenges the industry faces.

As highlighted with these two very different goals, the solutions presented not only shift the economy closer to becoming our ‘ideal’ economy, but they could also help solve the current problems. By having the goal in mind, we also have removed some options from the discussion because they do not help us to achieve our goal.

I believe that asking the right questions is key to solving Barbados’ structural problems so let’s start the discussion. What is your vision for the ‘ideal’ economy of Barbados?


What level of international reserves is enough?

Over the last few months we have all listened to the robust debate between our politicians and other interested observers regarding whether Barbados is holding an adequate level of foreign exchange reserves. While the debate has been useful, there is an implicit assumption that the international benchmark of an adequate level of international reserves is a good indicator. That may not be the case.


Why do we care about the reserves?

Before we can debate the benchmark, we need to understand why we even care about the level of the reserves. A simple way to think of the reserves is as a backup foreign currency savings account. Consider this simple example:

  • Barbados receives money from selling its goods and services to other countries or when other countries invest in Barbados (foreign exchange inflows);
  • Barbados has to spend money on goods and services it purchases from other countries or when it invests in other countries (foreign exchange outflows);
  • If it spends more than it earns, it needs to make up the difference with savings (reserves).
  • If you run out of savings, how will you pay your foreign currency bills?

In fixed exchange rate economies like Barbados, international reserves allow the central bank to make the implicit guarantee that it will be able to convert local currency to foreign currency on demand.  Think about what would happen if the Central Bank of Barbados could no longer fulfil its promise to exchange two Barbados dollars for one United States dollar.  This would mean that in order to purchase materials from abroad, you would have to obtain foreign exchange from a foreign exchange dealer and pay whatever price the dealer demands for foreign currency.  In these circumstances, the market exchange rate would deviate from the pegged rate – i.e. the market exchange rate would no longer be 2:1 – forcing the authorities to abandon the peg. When you consider that we import almost all of what we consume, instability in the foreign exchange market would have wide-reaching consequences.

It is therefore not difficult to understand why so much emphasis is placed the international reserves in fixed exchange rate economies.  It indicates to investors (both local and foreign) whether the peg is sustainable, and therefore the risk of exchange rate-related losses if they were to invest in the country.


 Why it is important to have the right benchmark level?

I like to think of the current reserves debate as similar to steering a ship while tracking the development of a hurricane. We rely on our meteorologists to keep a close eye on every patch of cloud in the sky and watch how it develops. We expect to get warnings long before a harmless cloud becomes a hurricane. We expect to be told how quickly it was developing and when and where to expect it to hit. We may not fully understand how they do it, but we assume meteorologists have thresholds that they compare developing storms to.

But what if their thresholds are wrong? Or, what if their thresholds only signalled an approaching hurricane when we were already in the midst of it? There is nothing wrong with having thresholds. The challenge is obviously if the thresholds are adequate and give us enough time to prepare. That is the concern I have about the internationally accepted benchmark of an adequate level of reserves. I think we all accept that the country’s external position is at a very vulnerable point, but is it too late now to do anything about it?

The current benchmarks

So what level of reserves is needed to maintain the peg and support investor confidence? Economists have two popular indicative rules of thumb.  One of the most quoted in local debates is 3 month or 12 weeks rule: reserves should be able to cover 12 weeks or 3 months of projected imports.  It is important to know that there is no statistical justification for this ratio; it is just a figure that has become a focal point for economists.  It is also subject to criticism as it ignores financial flows and focuses solely on the flow of goods and services. In a country like Barbados, this is an important failing, because while we import more goods and services than we export, we have historically attracted more capital than we have been sending abroad.

Another widely used indicator is the ratio of reserves to broad money (or the amount of printed currency as well as checking and savings deposits in a country). Consider the basic commercial banking model. Banks accept deposits from people that have surplus funds and then lend these funds to people that are in deficit and need to finance their activities.  Banks therefore never have all of the deposits that they have collected in their vaults. The model works, since only a fraction of depositors would demand access to their funds on any given day.  The same thing applies with reserves. If you think of the amount of money in a country as the deposits, only a fraction of these ‘depositors’ or holders of local currency would want to convert their local currency into foreign currency on any given day.  The benchmark considered adequate in this instance is that reserves should be around 20 percent of broad money, i.e. only 20 percent of the money supply would need to be converted into foreign currency in the short- to medium-term, whether it is for goods purchases or investment purposes.

 An alternative benchmark level

While I think that these two ratios are useful, I disagree with the benchmark levels that have been treated as sacrosanct in local debates because they ignore the many other factors that need to be considered when determining the benchmark in the first place. It’s similar to if we are only informed of an approaching hurricane when rain has already started to fall and winds are already creating 20-foot waves. Our international reserves benchmarks are far too low to be meaningful as a signal of distress.

Now let’s go back to the question we raised in the beginning: are the level of reserves Barbados currently holds adequate?

Looking at the benchmark indicators, you could conclude that the level of reserves appears to be adequate.  The reserves currently cover 13.3 weeks of imports of goods and services and at March 2013 were 22% of broad money. In both cases the country is above the benchmark levels so we should be in the clear. However, the story is a bit more complicated.

Consider these three facts:

  1. Small states are more vulnerable to natural disasters and therefor require larger amounts of foreign currency to finance recovery efforts;
  2. Foreign currency flows quite freely in and out of the country and we have limited ways to control it;
  3. We have a very large public sector that drives a large proportion of the country’s foreign currency consumption.


This would imply that we would need to have a larger cushion than countries that do not have these characteristics. We have derived an ideal target of 22 weeks for the ratio of reserves to imports (see the recent paper where we determined the optimal level of reserves based on a statistical cost benefit analysis for more information)[1]. This is almost twice the rule-of-thumb of 12 weeks and is a direct result of the high probability of natural disasters in small states.

Even this target, however, should not be viewed as sacrosanct. We found that small states that were able to implement a prudent government expenditure management framework would be able to hold a smaller stock of reserves, without leading to a disruption of normal import and export activities or any negative impact on short- to medium-term growth. Though the latest Central Bank of Barbados press release showed declining expenditure levels, the fiscal deficit was still estimated at a relatively high 8% for the 2012/13 fiscal year.

In short, it is important that we monitor the level of international reserves in countries such as Barbados. It is equally important that we independently determine what level is the minimum that we need to comfortably meet our obligations.


[1] Moore, W. and Glean, A. (2013), Optimal Foreign Exchange Reserve Holdings in Small-Island Developing States, presented at the Central Bank of Barbados’ Annual Economic Review, Bridgetown.