Aug 22, 2020

Akamaisation

(This post originally appeared at Swarajya magazine here

During the 1990s, computing faced a unique ‘memory wall problem’ due to increasing processor clock speeds but relatively slow memory speeds. While processor speeds increased at 50% year on year (Moore’s law) memory speeds increased only by 10%. Imagine a talented and quick chef bogged down by a slow assistant, who takes longer than usual to fetch ingredients from the pantry. The solution was found by using a ‘cache’, a smaller but faster memory placed nearer to the processor, thus avoiding multiple trips to the larger and slower main memory. This was like the chef getting a shelf that stocked all recently used or frequently used ingredients closer to his table. Today we have multiple caches at multiple levels in any given computer. Similar problem of earth-size arose when digital contents were to be served to multiple users across the globe. Imagine Netflix maintaining data centers only at US and trying to service the users across the world. The geographical distance that each data packet needs to travel would be huge and may clog the network affecting speed and user experience. The solution is again in a form of caching created by ‘Content Delivery Networks’whereby proxy servers and data centers are maintained nearer to the users, optimizing the content travel over the network. One of the largest firms in this field is Massachusetts headquartered Akamai which roughly covers a quarter of the web traffic today. Akamai is present wherever the users reside and maintains around 3 lakh servers in 136 countries and claims to have 85% of internet users within single network hop to an Akamai server. 

 

Going local as an optimization technique was a natural development in the above cases. This may also apply to the world of manufacturing. Supply chain management is known to use such optimization techniques while deciding on warehouse location. But when it came to manufacturing, scaling up at one location was usually considered an ideal way, especially in the recent decades. This is now changing, and China centered global production models is on the verge of being altered where production would take place in multiple geographies to cater to local markets. It will be the next logical step for production firms to survive in a de-globalizing world. Geopolitics, advancements in automated production techniques, declining appetite for globalization, rise in border barriers, and rogue behavior of China are the precipitating factors that would push this movement for change. We cannot term such optimization by production networks as ‘deglobalization’. A parallel term is ‘Akamaization’ – a solution where the firms go local in order to remain global.  Akamaization would arise in the beginning due to multinationals adjusting their international supply chains to the reversals in globalization by leveraging on Industry 4.0 advancements such as automated manufacturing, robotics and IoT. In the second stage, Akamaization would include growth of organizations that design their manufacturing processes to be agile and Akamaization friendly. 

 

The good years of globalization saw concentration of production in labor intensive countries. This was possible due to falling tariffs, technological developments, and logistics optimization such as containerization of cargo. The scale of concentrated production decreased costs and the decreased barriers to trade alongwith cheaper logistics ensured that geographical distances didn’t matter. The data suggests that this model peaked sometime around 2012-13, few years after the financial crisis.  Both in terms of merchandise trade as percentage of GDP as well as total merchandise trade as measured in current US Dollars terms has stagnated (or declined) after the recovery following financial crisis.  One may say that the world reached ‘peak trade’ during the first half of 2010s. The decades leading to peak trade were marked by growth and concentration of manufacturing in China and nearby countries including Taiwan, Korea and ASEAN. 


Akamaisation: Merchandise trade as percentage of GDP

 Fig: Merchandise trade as percentage of GDP - Source: World bank

 



Akamaisation: merchandise trade current USD

Fig: Merchandise trade in current USD - Source: World bank

 

There is no single explanation for peak trade. The most likely explanation is that the factors that drove the growth had run their course and there was not much that could be done further on. However, the decline starting thereafter and the expected decline in future would be of a different nature.  It’s easy to call it de-globalization, but for the adaptability of the firms that run the international value chains. 

 

In the last few years, after decades of embracing globalization, developed countries finally woke up to the unsavory side effects. ‘Hyper-globalization’ – a term used by Dani Rodrik – of the last three decades where capital as well as value chains cut across borders, is not proving to be a sustainable proposition in the long run as the free trade of hyper globalization type creates a losing side. The political setup of the losing side can’t explain away the job losses with consumer welfare or Ricardian theories. Free trade had peaked before the emergence of the protectionist political leaders who called it out. While the globalized model with zero barriers suit the multinationals, the politicians and policymakers increasingly realize the bad effects on domestic job growth and general prosperity. The backlash by the increasingly frustrated population is partially responsible for emergence of protectionist political leaders. Similar sentiments may also be seen in the prevailing mood of going local slogans such as ‘Make America Great’ and ‘Atmanirbhar Bharat’. While the free trade economists may rue the developments and sermonize about unlearnt lessons from history, it provides no solace to the multinationals who operate the international value chains and who need to explore alternative models to remain in business. Deglobalization by disengaging or shrinking their markets and business is not an option for the globalized firms. That’s where Akamaization steps in. 

 

Akamaization of a global firm involves higher level of integration and interdependence in terms of management, design, and product development while maintaining a network of production setups catering to limited geographies isolated by rising trade barriers. Imagine a global cell phone manufacturer rolling out new products every year but choosing to make similar models in different key geographies with significant market shares. Most components may be sourced locally to avoid border crossings. The manufacturing setup would cater to the domestic market, and export to those nearby markets where the market share is not significant enough to warrant setting up an independent manufacturing base. The optimization problem for an Akamaizing firm would involve variables such as direct and indirect cost of crossing borders, geographical distance, market size, investment and business costs etc. Akamaization may also happen due to geopolitical pressures. For example, Taiwan’s TSMC, the largest chipmaker in the world, has announced that it would be setting up manufacturing in US in response to Trump administration’s pressure to move critical technology back into US. 

 

For policymakers Akamaization starts with attracting ‘screw-driver’ technology where assembly operations would move in first, followed by indigenization of components. This can be engineered by carefully planned tariff and non-tariff barriers across the value chain starting with higher barriers for fully assembled products and moving down to smallest of the components in a graded manner.  A success story for India is the mobile phone manufacturing which got an upshot due to the border duty tweaks on mobile phones during 2016-17. Apple’s decision to set up assembly plants in India and further plans to increase local sourcing of components is an example of Akamaization. The growth of mobile phone exports from India may be seen below. 

 

Akamaisation: Export of mobile phone and parts from india


Fig: Export of mobile phones and parts from India in USD Billions - Source: UN COMTRADE

 

Now the Ministry of Electronics and IT is incentivizing production of electronics and mobile phone components in India through various capital investment and manufacturing linked support plans. This may further accelerate the trend shown above, not only for exports but also in terms of decreasing dependence on imports of electronics hardware. This support shall go hand in hand with border barriers for import of the components. It must also be noted that there would be limited scope for export led growth in an Akamaizing world in future, a factor that policymakers must consider while intervening. 

 

Akamaization would not remain restricted to fields such as automobile, electronics or technology products alone. Even in fields such as garments and footwear, there is a chance that the ever-changing fashion trends and the need to be nearer to the market may drive Akamaization. The high labor input in some of these fields may delay Akamaization, but beyond a threshold and with rising border costs, the cost benefit analysis shall tilt towards manufacturing near the markets, starting initially with the high-end segment. As the manufacturing technology improves with advances in technologies such as 3D printing and robotics, even the fields that were considered offshoring friendly would be relooked at. When Akamaization touches labor intensive fields, it may affect the employment potential of those fields. This would be a challenge for labor rich countries like India which should explore upskilling the future generation to cater to altered needs of the future industry. 

Akamaization also has the potential to divide world into groups based on mutual trust on intellectual property (IP) front. Like minded countries that are trusted with IP and show greater regulatory coherence would bond together. The experience of US and western firms with China has shown the importance of IP protection and preventing corporate espionage. This may build a new world order based on emerging alignments where trade and geopolitical interests converge, breaking away from the existing multilateral arrangements.  Firms would be more comfortable Akamaizing within the group. 

 

It is important to understand that Akamaization would happen under a certain set of conditions. These conditions include a significant market size and right policies for setting up manufacturing in the country. If the market size decreases, or if the border resistance rises to unsustainable levels where it no longer makes sense to transact anything across borders, firms may give up on the market over Akamaizing. The same goes for steps towards greater ease of doing business which facilitates easy Akamaization. It’s easy to miss the balance, and that would bring back the free trade economists who have been predicting doomsday since the time India tried to fix the distortions in trade arising out of poor policy choices in the past.

 

 

 

 

 

 

Apr 8, 2020

Would more COVID testing uncover more COVID positive cases?

It is alleged that India is not testing enough. The number of tests per million population in India stands at around 100 when compared against countries like USA which stands at 6300, or Italy which stands at 12000, or Switzerland which is around 19000 tests per million population. 

So, is it that because India is testing lesser number of people per million population, India is reporting less number of cases? Would more testing lead to uncovering of more cases, increasing India's infected count?

Ashish Chandorkar, in this Swarajya magazine article disagrees. Rightly so. He states: 
"What does the data tell us?

In India, out of the 100 high risk cases being tested for COVID-19, only 4 are testing positive.

Of those testing positive, roughly sixty per cent are either individuals who traveled from a foreign location or were part of the single source Delhi Nizamuddin congregation.

The positive cases per 100 tests is 17 in the USA, 19 in Italy, 23 in the UK, 24 in France and 37 in Spain.

What does this tell us?

That in spite of the fact that the USA has done more than 16 lakh tests and Italy has had more than 6.5 lakh tests, there is a need for more testing as a large fraction of people are turning to be COVID-19 positive.

Testing in itself is not a panacea to all the problems being associated with COVID-19. Testing for testing’s sake is of no use. Testing should be done for identification and for quarantine."

While that was my hunch too for long, and I agree with Ashish, I still thought of running the numbers quickly and plotted a scatter bubble for all data available for all countries across the world. Here it is: 
(A note here: We are talking about percentage of cases, not absolute numbers. While more testing will lead to more absolute numbers, the percentage positive would remain mostly a constant)


image for Test per million capita Vs Percentage positive cases COVID testing
COVID Testing - Would more testing help?


It's always tricky to explain a plot unless the intuition autoclicks for the onlooker. 
On the X axis you have the Tests per million on log scale. On the Y axis you have % of tests that turned out positive. No country reports that all tests conducted were found COVID positive. France comes close to a point where 1 in every 2 tests return positive. In India for every 25 people tested, one returns COVID positive. For USA it is one positive in five tests and so on for all countries as one can see. 

One may note that there is no correlation between percentage detection and per capita testing. It means that wider coverage won't necessarily lead to more detection of cases. To be sure, I have also plotted a best line fit which has almost no slope (slightly negative to be precise). This can be seen below (with a pinch of salt due to the Rsquared value): 

image of COVID test success percentage vs more per capital testing per million population
COVID Testing: Relation: Test success percentage(Y) vs More Per M capital Testing (X) 

So, the data supports the theory that more testing won't help unless we have more infections in reality.   By testing more, we will just go where Iceland is currently in the first plot. They are testing huge numbers with the resultant same level of success that we have in terms of percentage. We shall move on X axis without any change in Y. 


The data so far for India says that the infection level (on per capita terms) is very low, and there is not much transmission on the scale seen in other countries. When we juxtapose against the fact that India has been testing high potential cases (foreign returns, NZ contacts, first level contacts etc.) and still returning negative results, it bears all the more testimony that probably our level of testing is not to be blamed for low numbers. And if that is so, we have reasons to cheer. Either the lockdown has worked, or other factors such as malaria resistant people, hot/humid weather, demographic profile, inbuilt resistance etc is at work. 

The plot would also be interesting to watch over time for another reason. If the position of a country on Y axis doesn't change over time, or with increased testing, it would prove that a country exists with a certain tolerance for the infection/spreading which is a function of factors other than testing. Once unmasked, this factor would tell us what might further help. 

Above plot were generated on 8th April 2020. 

Updated figure as on 16 April is below: The point has just moved right. I won't be surprised if it goes and touches Iceland one day. 

COVID percentage detection has not changed for India - As on 16 April 2020



Updated figure as on 26 April is below: The point has just moved right as we have increased the number of test per million from around 200 ten days ago to around 450 currently. It's moving horizontally as expected. 

figure of COVID percentage detection has not changed for India - As on 26 April 2020
COVID percentage detection has not changed for India - As on 26 April 2020

Updated figure as on 03 May 2020 is below: One million tests completed for India. A Milestone reached today. The point has just moved right as we have increased the number of test per million from around 450 a week ago to around 800 currently. It's moving horizontally as expected. 


Updated figure as on 10 May 2020 is below:  The point has just moved right as we have increased the number of test per million from around 800 a week ago to more than 1000 currently. It's moving horizontally as expected. 


Updated figure as on 31 May 2020 is below:  The point has just moved right as we have increased the number of test per million from around 1000 in the previous figure to more than 2800 currently. It's moving horizontally as expected. However, there is a slight bump in Y axis by roughly around a percent (4% moves to 5%)

Updated figure as on 20 July 2020 is below:  The point has just moved right as we have increased the number of test per million from around 2800 in the previous figure to more than 10000 currently. It's not moving horizontally as expected and for the first time I see a slight vertical movement too. From 5% it is gone upto around 9%. Still much lower than 50% that some countries demonstrate but a movement up nevertheless. I would change my mind about the theory if it goes above 15%. 


Add caption



(Note: All data taken from Worldometer. I have placed the code for generating the labelled bubble plot at https://github.com/tirumalakv/COVID19plot.git to tinker around if you are interested. The plot with the regression line was generated on MS Excel)