Resilient Profitable Dairy

Headlands Conference: “Resilient and Profitable Dairying.”
A paper presented by B J Ridler (GSL).

(Resilient: flexible, quick to recover).
B J Ridler (GSL) W J Anderson (Massey University).
This term means little when searching for policies to implement on farm.
If we consider the reasons why a system may be resilient we must return to the basic resources and efficiency of allocation over a wide range of circumstances.
Such analysis can then define a set of criteria whose application over time can be labelled as “resilient”.
But such a descriptive word does not move us any further towards understanding what factors create such a state, where we should concentrate our efforts to improve the system nor why we should do so (does it lead to more money: how much more?. Less risk; how do we compare this?)

If it is a physical resilience, the basic resources of the farm (land, climate, and location, stage of development, stock and management) all integrate in an efficient manner. These circumstances can be achievable across differing aspects of resource quality, but the probability of this occurring reduces as the entropy (the degree of disorder or randomness of the constituents) of the system, increases.
Hence the more marginal the land or resources are, the more disordered they will be and the more expensive it will be (initially in terms of energy) to ensure such resources remain at, or improve, to a more efficient (less disordered) state.

This statement of just one of the factors associated with the laws of thermodynamics should prompt a rejection of the current flawed methods of comparing farms using benchmarking and KPI’s across farms and farm systems and a return to a more applied quantitative analysis of each individual farm system. The entropy difference between each farm system is different and by definition not possible to compare accurately.

So when there is talk of “resilience” we should cut to the underlying factors which determine the likelihood of minimising entropy within systems.
Quality of the base resources form one constituent of efficient systems and economic factors invariably forms the other when successful businesses are required.
The requirement for low debt levels, strong cash flows, low fixed cost outflows including interest and depreciation must be combined with high animal and pasture efficiencies.
These must then be combined to produce an overall system where all resources are allocated efficiently within the constraints imposed i.e. low entropy.

At farm level such a combination will include minimising fixed costs and the associated “fixed variable costs” (those costs that are variable in nature but occur due to a fixed commitment that requires concomitant expenditures to be made. Example: a covered feed area where additional feeds to either feed more bought in feeds or improve current per cow production requires additional machinery and maintenance for which additional labour must be employed in order for the added infrastructure to be used). True variable costs are minimised when constraints within the system are identified and can then be largely avoided.

All this becomes possible when efficient agricultural systems are implemented.
Obviously, high input farms involve increasing most of the factors that decrease resilience whereas low input farms move towards the factors that improve resilience in all circumstances.

However it is clear that such terms as “resilient” are an excuse to avoid the work required to quantitatively identify then define the constraints within specific farm systems. It is not until this quantitative work is completed that we can become more objectively definitive of just when each specific farm may alter from one state to another (the “tipping point”), how rapidly that may occur (under what circumstances) and the impact this will have (sensitivity to change).
This requires application of farm management principles such as production economics. Without this, the marginal changes that occur with additional inputs and the selection of alternative resources (substitution) on the basis of overall systems efficiencies will remain in the realm of “reading the tea leaves” from data maps and be justified by coining subjective and emotionally qualitative terms.

When farms are monitored, detailed data are accumulated on daily, weekly or monthly periods. These are the only useful data. Once averaged, the figures that result are now once removed from the actual farm and such data now lack the detail required to quantify change and identify constraints.

Rather than use detail and quantitative analysis, the simpler method has been to gather data then corrupt its value though averaging, creating ratios of illusionary efficiencies then using these to compare across different farms.

The harsh reality is that identifying and understanding how best to overcome known constraints, exploring where and when the “unknown” constraints may occur and assessing the risks (and costs inherent in such risks) of these to the current system is all part of providing a stable and profitable system.
If it is not at the peak profitability, the system will also likely be performing poorly environmentally (Anderson 2010; Ridler 2010; Ridler 2014).

GSL has used these principles to construct a full farm systems model using factors of production to set up a final analysis using linear programming.
Such a system ensures that entropy does not alter from one year to the next. The energies used must balance and return the system to the same level at the conclusion of each run to ensure a pure comparative analysis for any change being investigated. This ensures a “pure” mathematical solution which resolves all factors of production in a balanced and integrated manner.

All other models may “optimise” various parts of a system (components) but then the operators “jam” these components into their perception of what the system should be. Invariably the optimised components do not then combine in a manner that enhances the overall system. GSL takes over the role of overall systems optimisation, but can be guided by operators -or bludgeoned by operators. The difference here with LP is that if this approach exceeds specific biological or economic limitations, no solution is returned.

The distinction between biological and economic efficiencies is able to be explored by such a self-balancing routine.
It is difficult otherwise to illustrate and to convey to farmers, vets, consultants and other agricultural people that biological and economic efficiencies are often a long way apart.

What we must also realise from the First and Second Law of Thermodynamics is that attempting 100% efficiency in any process is an expensive and on-going challenge that is, economically, rarely worth achieving except where personal objectives override all other factors.

Averages and Ratios:
With the advent of more sophisticated and powerful computers (plus programs beginning with Lotus 1 2 3 and now Windows Office Excel) has come the ability to process data quickly and efficiently, but not necessarily sensibly. Thus such data manipulation comes with increasing reservations over the effectiveness or efficacy of much of the data crunching that is now undertaken.
With the ease of measuring data today, increasing amounts of farm data is being produced, accumulated into spreadsheets, averaged, formed into ratios then used to compare between farms. Throughout this process, the distinctly different factors of one farm’s production system are somehow assumed to become equal or irrelevant when averages, ratios, benchmarks and KPI’s are calculated from CLUSTERS of farms using MERGED data then compared to establish correlations BETWEEN farms which, in reality have quite different characteristics.
“Shotgun” patterns of data can then be happily manipulated to provide correlations and these then discussed at length, but all this is in reality, a farce.

Candler (1962) noted that;
“The apparent simplicity of this procedure, which can be followed without any a priori hypothesis as to the key to successful farming, should make one suspicious.”

(This was decades before the power of computers made the process even more suspect!)

And if the concept of entropy is understood, (that in any system there is a unique degree of disorder or randomness of the constituents) it should be realised that comparisons between farms where such randomness abounds and cannot be accurately measured, strongly supports Candler’s suspicions in using any data to compare between farms.

There may be some defence of such artificial measures if they serve to motivate managers to observe and attempt to understand their systems better, or perhaps where in the early stages of development, the average costs are clearly less than the marginal costs. The problem here is that this applies to few farms in New Zealand but many farms in other parts of the developing world where competition to our products is increasing rapidly.

Diminishing Returns:
If we look at a diminishing returns curve (Appendix 1) we can see or calculate where TR=TC (productionists) or TR/TC or AC=MR (most farm models) and where MC=MR (GSL).
(Marginal Cost MC; Marginal Return MR; Total Revenue TR; Total Cost TC; Average Revenue AR).
If we average a curve, where do we stop adding resources as all we have is a straight line?
If we have some data to establish a diminishing returns curve we know that each point corresponds to a different figure. When the figure begins to decrease (about 30kgN use /ha Appendix 1.) maximum return for each kg N used has been achieved but at about 43 kg N use/ha, MC=MR and the highest profit from using N will be achieved. After this point, although the average revenue of the nitrogen used is still more than the return and will be until about 60kg N/ha are used, the profits already made are being eroded away (MC/MR column).

Invariably other models all use average data that has cost averaged over total production,   NOT the marginal cost of the marginal production.
Such models will not cease inputs until the operator perceives AC = MR or perhaps where AC=AR. This is a point where for example, additional feed use is about 25-50% more than GSL due to the difference between marginal and average costs (depending on the factors being examined.)
So of course this means more inputs are added at a marginal cost greater than the marginal return which reduces profit but adds to N leach.
Yet many within the Dairy Industry refuse to acknowledge this fact which has been proven empirically by advisors (Alison Dewes 2014), the GSL model (Pellow 2013) and many more independent farmers. (The word “independent” is used as the push to increase production within the dairy industry has created “System 1” thinking -see later reference- in most dairy farmers).
If we use MC/MR methodology, inputs cease earlier, costs are saved and N leach reduced as bought in feeds are firstly reduced (MC>MR), a reduction in herd numbers follows and profit improves as there are less variable costs (feeds) and fixed costs associated with cows and feeding.

Ferris and Malcolm 1999 have stated this concept clearly:
“Farm businesses are usually characterised by average total costs (Fig 6) which initially decline as production increases and overheads are spread over more output. Average variable and total costs may then be relatively constant over a range of output, and can increase if inefficiencies arise from increasing size of operation. Marginal costs tend to increase as output increases, as diminishing returns to variable inputs sets in (as shown in Fig 6). Maximum profit in the short term is made at the production level where marginal costs of extra output is almost equal to price received per unit of output.”

Ferris and Malcolm(1999)

Through the early stages of agriculture in any country, easy gains are made to production as better species, fertiliser and animals are introduced. This means that the average costs of running the business have more influence than the additional (marginal) costs of adding resources. This is the LHS of the Figure 2 Graph.

Production increases may continue to occur at a marginal cost close to the average costs which results in fairly constant returns.
But when a farm has exhausted all the easy gains (productivity gains may still exist but are too often confused with “production” gains) any further production increases are gained by adding more cost for less return than the costs of the inputs. The marginal costs are now greater than the average costs to run the business and when the marginal return becomes negative, the business immediately begins to make losses on these parts of the business and this erodes overall profit.
If the MC vs MR on each component of the business can be calculated, the use of capital becomes efficient and productivity gains made by comparing options, identifying constraints, then selecting which constraints to overcome first (and at what cost/return) to increase productivity. It is highly unlikely that all parts of a biological system will be at top efficiency when the overall system efficiency peaks.
In almost all cases now in New Zealand, the marginal costs are doing the “heavy lifting” (Fraser 2014) yet the treatment of detailed data (and models that use that data) corrupt the detail required to extract any marginal values. Without this, any models are only as good as the operator’s best guess and lack the ability to provide outcomes that challenge the perceived wisdom of entrained thinking.

Without the ability for a proper comparative analysis based on a closed system of known resources which must balance, include the ability to substitute based on marginal costs and returns (MC+MR) and by this process, identify the timing and limitations imposed by constraints specific to the system, any farm management “analysis” becomes a self-fulfilling parody of a farm where solutions can be whatever the operator desires or, increasingly, what the project sponsor requires.

Any quantitative research and empirical application of scientific principles should discourage the use of emotive or descriptive words such as “resilience” in preference to quantifying the factors that allow any farm to reduce entropy, improve the balance of resources within a system and therefore become simpler to manage and make profits from.

The tendency to produce simple terms and thinking to describe complex factors is described as “System 1 thinking”. (Daniel Kahneman 2013).

System 1 thinking is rapid, instinctive, and subjective, uses links to previous patterns and is invariably inaccurate or just plain wrong. (Many large and expensive projects proceed on System 1 thinking.)

System 2 thinking however requires thought, quantitative analysis and an approach that minimises the illusion of validity. This is far more difficult for humans to process initially but is far more accurate and almost always correct. (Many large projects would be discarded if System 2 thinking prevailed.)

What is also clear is that averages, ratios and associations between components constructed from such manipulations of data are concealing the point where resource use must cease on the basis of marginal cost and return.
This is detrimental to the environment and the wider agricultural community.

GSL model discussion points.

A confusion between production structures rather than SYSTEMS.

Confusion between BIOLOGICAL & ECONOMIC efficiency.

Creating this confusion is the fact that:
Averages are used rather than actual (real) data.

The use of these averages then further disconnects the data from reality.

This disconnect is allowing incorrect policy to become embedded at all levels in New Zealand.

The biggest opportunities for improved profit in NZ agriculture are in areas that current analytical methods cannot detect.

Resource allocation linear programming model.

Methodology to set up and run LP allows data “gaps” to be filled.

Concept of diminishing returns is important when dealing with resources.

Average vs. marginal change to determine the “tipping point” for all inputs.

Detail is required to identify the tipping point.

Averaging destroys detail and encourages waste of resources.

This results in poor outcomes for both the business profit and the environment.

Positive Economic and environmental outcomes almost always coincide.

Models require transparency for farmer credibility and acceptance.

No “one result” is the answer to any complex systems question.

Risk and objectives need to be assessed.

GSL model.

New generation modelling.
Concept tested on real farm predicting outcomes of alternative management system. (No. 4 Dairy Farm Massey University).

Latest model uses new technology and methodology. “Road tested” on Lincoln University Dairy Farm 2010 to revise the dairy farm system which had “plateaued
New plan from GSL modelling implemented 2011 and results were almost exactly as predicted by 2 years later. (Pellow et al 2013)

Combination of new technologies and programs.

LP is ideal for complex systems solving but it is the conceptual thinking required to visualise and articulate the system that is the real problem.
Most “Optimisation models” do not optimise full systems, just partial sections.
For correct production economics and accurate bio-economic systems comparisons, ability to substitute resources on the basis of MC and MR is required.
This allows constraints to either input OR output functions. This allows the model to optimise all resource use as N leaching limits are imposed as an example.
Such a process allows the optimisation to substitute or remove resources on the basis of best economic and farm system resource allocation outcome from decreasing N limits.
Such an outcome is the result of not one but many hundreds of thousands of calculations where the model moves closer to the best solution by means of a marginal analysis process made possible by the LP process.

Lack of systems knowledge and attitudinal bias are major constraints for GSL’s widespread use.

GSL has been widely used for specific projects:
Economics of N reduction on dairy Farms (LUDF DairyNZ and Regional Councils).

Management options for N reduction on specific dairy farms.

The value of specified irrigation systems for large multi system farm enterprises.

New management options for “Focus Farms” with regard to improving economic outcomes and capping emissions (LUDF 2010-2012, Southland Demonstration Farm 2012; Massey University No1 Dairy Farm 2013)
In this way it is a model whose future predictions and farm systems changes have been the most widely scrutinised of any farm systems model in the world.
Despite some detractors, it has not been found to have predicted wrongly.

Anderson, W.J; Ridler B.J: The effect of increasing per cow production and changing herd structure on economic and environmental outcomes within a farm system using optimal resource allocation. Proceedings of the 4th Australian Dairy Science Symposium 2010.

Candler, W and Sargent, D. Farm Standards and the theory of production.
J.Agric.Econ. Vol 15 No.2 Dec 1962

Dewes, A.: Economic Resilience and environmental performance of dairy farms in the upper Waikato region. Master of Science thesis. University of Waikato 2014

Ferris and Malcolm Agribusiness Perspectives Papers 1999 Paper 31 ISSN 1442-6951
Sense and Nonsense in Dairy Farm Management Economic Analysis
Alexandria Ferris and Bill Malcolm
Department of Food Science and Agribusiness
Institute of Land and Food Resources

Fraser, Ridler, Anderson: Intensification of the NZ Dairy Industry Paper NZARES Nelson 2014
Daniel Kahneman. Thinking, Fast and Slow. Penguin Books. April 2013
( or (

Pellow,R. Steve Lee, Alister Metherell, Roy McCallum, Jim Moir, Ants Roberts, David Wheeler. Assessing the impact of input choices within overseer® (v6) on the modelled N losses to water for Lincoln University Dairy Farm (LUDF) 2013

Ridler,BJ.; Anderson WJ.; Fraser P.; Milk, money, muck and metrics: inefficient resource allocation by New Zealand dairy farmers. NZARES 2010

Ridler,BJ.; Anderson WJ.; Fraser P.; “Win/Win. Improving farm profit and the environment through the application of Farm Management principles. NZARES 2014

Appendex1. Response to Nitrogen applications. "Diminishing Returns Curve."

Productionists apply N to about 92 kg N/ha as Total Revenue-Total Cost is still positive.
Averaged figures allow application of N until about 63kg N/ha as Average Revenue is about same as added unit cost.
But the correct calculation is that application should cease at about 45 kg N/ha when the added cost of the extra unit of input exceeds the return from that input. $1.64 or Marginal Cost = Marginal Return (MC=MR).
But to calculate this correctly requires marginal analysis which is difficult when averaged figures are used in an input/output format. The GSL model allows marginal changes for nitrogen applications to be compared for each 2 week application period in order to choose how much to apply and when. This depends upon the economics of the added feed within the farm system being evaluated.
The same applies for bought in feeds (see below: Appendix 2 and 2a)

Appendix 2: Cost of bought in feeds.
Bought in feed combination costs $300 / tonne landed on farm at 90% DM. This is 33 cents per kg DM. Also the mega joules per kilogram of dry matter (MJME/kgDM) will vary in each feed as will the crude protein and other factors.
Feed out costs of this will vary depending on quantities and facilities. Assumed a “budget” tractor and feed out wagon on farm with dry area to feed when necessary. Costs for this will be from 4 cents upwards (machinery, labour time.) If higher amounts fed higher capacity machinery and a feed pad system plus in shed feeding will be required cost will be 7-10 cents per kgDM fed.
33 cents PLUS feed out costs = 37 cents /kgDM
Utilisation: If a reasonable dry paddock or area this will be about 85% utilisation so the costs will now be (37 cents x .85 utilised =) 44.5 cents per kgDM utilised.
Feed pad etc. improves utilisation but the cost of the facilities have increased the feed out costs. Calculations can vary but now 33 cents + feed out costs of 8 cents x 95% utilisation = 43 cents.
But this feed is still not equivalent to pasture so add another 10% (10.8 MJME vs 12MJME/kgDM) so about 46-47 cents/kgpasture DM equivalent.
Each step increases the real input cost compared to the original figure for the bought in feed.
Any cows fed on such bought in feeds are marginal cows and other fixed costs to run these cows must also be added to this sum. These include added labour (additional 106 cows will require ½ FTE) animal health and breeding expenses plus some additional per cow for added farm maintenance. DairyNZ Economic Survey 2012-2013 page 44 Table 5.8 per cow costs for these additional cows will total about $400 per cow. (This figure excludes the cost of additional capital required for the cow and part replacement, increased milksolids shares, plus any additional machinery and infrastructure.)

Appendix 2a: Varying costs per cow depending on per cow production.
Depending on the production per cow, the intake of feed at “normal” dairy farm pasture qualities (varying from 11.5-12.0 MJME/kgDM in a seasonal pattern) will be:
5480 kg DM for 330 kgMs cows with 25% replacement rates and about 5 years in the herd. (If there is a desire to make a ratio of this, this becomes about 16.6 kgDM/kgMS).
5725 kgDM for a 400 kgMS cow with 25% replacement rates and about 5 years in the herd. (If there is a desire to make a ratio of this, this becomes about 14.3 kgDM/kgMS).
6020 kgDM for a 450 kgMS cow with 25% replacement rates and about 5 years in the herd. (If there is a desire to make a ratio of this, this becomes about 13.4 kgDM/kgMS).
From this a more accurate figure for Marginal cost for these cows can be calculated.
As the MJME of the cheaper bought in feeds is about 10.5-11.0 MJME/kgDM, the 43 cents must be equalised and this will make the cost of bought in feeds now about 46-48 cents per kgDM equivalent actually consumed by the cow.
Which makes cost of added feed per cow about (5600 x .47 =) $2600 per cow and the additional running costs per cow add $400 to total about -$3,000 per additional cow.
The additional MS in this case will be about 380 kgMS at $7 or about +$2,670 per additional cow. This leaves a deficit of about $330 per additional cow at $7/kgMS
DEFICIT $330 at $7/kgMS
OR $720 deficit at $6/kgMS
OR $1100 deficit at $5/kgMS
From feeding extra cows on bought in feed.
This does not include costs of additional capital, depreciation or maintenance.

However once again, the GSL model shows it is not as simplistic as this with the MC/MR varying with each additional cow depending on how the additional cow unit fits the overall system. In reality, adding cows up to, then past the “optimal stocking rate” where pasture plus true supplement (to fill a genuine pasture feed gap) balances shows that the diminishing marginal value of adding cows is less BEFORE the optimal point than the marginal cost increases which increase at an increasing rate AFTER the optimal point is exceeded.