Saturday, June 25, 2016

Statistical Freight Rates Predictions

The question that this paper intends to answer is whether freight rates can be predicted accurately applying statistical and network methods?
I had the pleasure of watching an online video lecture in Computer Science by Professor, John Guttag, at the Massachusetts Institute of Technology: “Using Randomness to Solve Non-random Problems”. He wrote a small program in Python to obtain the ratio of Pi using the old problem of randomly dropping needles on a drawing of a circle inside a square to obtain the ratio between the two drawings. This experimental geometric probability problem was first proposed in 1777 by Georges-Louis Leclerc, Comte de Buffon. He never obtained a good answer because his sample of trials wasn’t big enough or his method of dropping needles was poorly designed. A correct solution was given by Laplace in 1783 using binomial distribution: np (1-p) and sqrt (np (1-p).
Professor’s Guttag Python model simulated 20 trials starting with 1000 needles and doubling the number of needles each time to obtain a Gaussian distribution. As it turned out 16,000 needles provided a Pi ratio of 3.1413875, however, with 32,000 needles the ratio was 3.1457. While the program continued to double the number of trials, the trial ratio did not improve its variance. On the other hand, the standard deviation became smaller with each doubling of the needles from .012 and 16,000 needles to .008 with 32,000 needles. So, even when there was variance in the trial ratios throughout the trials the standard deviation got smaller every time where two standard deviations would be a very small fraction and a confidence level of 99% could be attained.
As stated by Lancaster University, UK: “In the 1940’s Ulam and Von Neumann suggested that aspects of research into nuclear fission at Los Alamos could be aided by use of computer experiments based on chance. The project was top secret so Von Neumann chose the name Monte Carlo in reference to the Casino in Monaco”. And so, experiments with randomness became better known as Monte Carlo simulations.
This begs the question of whether building a model of normal distribution using published liner shipping rates and comparing it with a model of random rates is accurate. The answer should be yes. We know from statistics that real life events have a normal distribution and when compared with a model of virtual reality there is only negligible variance between the two models.
With this in mind can freight prediction be made and plugged into a transportation network of nodes and edges where the nodes have all the attributes embedded or extracted from databases? This, of course, would be a challenging coding task but this paper proposes that such project is doable!
To begin with and in order to simplify building a model, only known variables in a rate structure should be considered. For example, a typical freight rate structure contains distance, dimensional weight and the rate per unit of cargo to be shipped. This model will not include the occasional variables of weather, accidents, labor strikes, port congestions and consider them as out layers. Also, a stable model is assumed for a length of time (one year) where fuel prices, route competition, crewing cost remain steady and there is fair play in the market between shippers and carriers.
Fair play in the market is the most difficult variable to predict because of its numerous components. Just as in financial markets the better informed party wins the day bearing in mind that better information empowers consumers in an impartial market.
Estimating the actual value of trade between countries is a strenuous effort not only for regulators but also for national institutions responsible for measuring the behavior of international trade and its effects on local economies. Among these institutions are the UN, OECD, The World Bank, The IMF and others but they all get their values as quantified by national governments, trade banks and others such as Global Insight, Platts etc.
The task of estimating the value of trade would easy if the same factors were applied globally in a linear relationship. However, this not the case, as the same unit of cargo will have a higher transportation cost from a landlocked country than from a developed country with multiple port facilities.  Also, asymmetry of trade must be factored in; in addition, competition in the route or the lack thereof injects itself between nominal and real prices.
One general measurement assessed by Custom Unions is the difference existing between clearance values from origin and destination countries. The problem with this measure is that Customs depends on declared invoice value to make the assessment of duties after discounting transportation costs. So, if a commodity is purchased Free on Board basis (FOB) or Cost and Freight basis (C&F) the difference between the two values most be transportation. Unfortunately, this is not the case as declared values are manipulated by buyers, sellers and carriers while Customs use tariffs to determine duties--transportation tariffs can be ambiguous in a deregulated market. And so, the real cost of transportation is at best, an estimated value for reporting institutions.
Moreover, Customs looks to documentation provided by the shipping company, as opposed to the documentation between the buyer and the seller. Information from the buyer and seller often contains estimated transportation costs or charges, while documentation from the shipping company contains the cost for the shipment. Notwithstanding, there is no precedent which specifically address the issue of whether terminal handling charges or other surcharges like fuel adjustment factor and value added services are to be taken into account when determining the actual amount paid for international transportation. The reason is that this additional cost factors sometimes are included in the basic freight rate tariff and other times is not as a way to achieve pricing differentiation between carriers.
One method used by some NGOs for estimating transportation cost in international trade is to fix a point for both origin and destination and assess the average increase or decrease in cost of commodities going through the fixed point in multiple directions. The Strait of Malacca has been used as a fixed point of reference for this purpose.
In conclusion, predicting freight rates using a combination of statistics and networks based on published rates should provide a probability confidence of 95% with one standard deviation from the mean in a basket of rates offered by other carriers in compatible routes is doable.
Alfonso Llanes
July 4, 2015
References:
John Guttag. 6.00SC Introduction to Computer Science and Programming, Spring 2011. (Massachusetts Institute of Technology: MIT OpenCourseWare), http://ocw.mit.edu

Kady Schneiter Utah State University Kady.schneiter@usu.edu. Exploring Geometric Probabilities with Buffon’s Coin Problem. Fall 2014.

Monte Carlo Simulation - A Brief History - Lancaster University www.lancaster.ac.uk/pg/jamest/Group/intro2.html

Maritime Operations and Logistics Data and Analysis

The World Customs Organization (WCO), established in 1952 as the Customs Co-operation Council (CCC) is an independent intergovernmental body whose mission is to enhance the effectiveness and efficiency of Customs administrations.


No comments:

Post a Comment