Tilburg University
On the compensator (Part II)
Merbis, M.D.
Publication date:
1983
Document Version
Publisher's PDF, also known as Version of record
Link to publication in Tilburg University Research Portal
Citation for published version (APA):
Merbis, M. D. (1983). On the compensator (Part II): Corrections and extensions . (pp. 1-20). (Ter Discussie
FEW). Faculteit der Economische Wetenschappen.
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
CBM
R
7627
1983
9
TflD;~' ;~1cTENE3UREAU
BiB~?"JTH~;~ K
K.?'1': I.~' ~ I'.- i,E
HOGESC'ri ~OL
i:~.QUtiG
subfaculteit ripr
Pf`A~1A~ofr~e1
~ C IiN~~iHU~uiiiiuiii
N 0 1 8 4 4~iii
REEKS "TER DISCUSSIE"
iuiui
ii
No. 83.09 januari 1983 On the compensator
Part II, Corrections and Extensions
1
-1. Introduction
In part I we introduced the matrix minimum principle (MMP) as a technique to solve the control problem, the filter problem, the LQG-problem, and the two-DM Nash problem. A formulation of the b1MP as a theorem can be found in appendix A. It appears that some results of part I need to be stated more precisely to over-come the difficulties in solving the Nash problem. Therefore, the filter problem will be reconsidered; the control problem which is in fact similar by duality is omitted, but can be treated along the same lines. Also the LQG-result can be ex-tended (section 3) for other representations. These results are needed for a better understanding of the sepa.ration property, which plays a crucial role in LQG-models.
2
-2. The filter problem revisited
For completeness, the model and the problem formulation will be restated briefly.
Model: xt~l -~t } M~t' xp E G(m,E) yt - Cxt f Nvt, vt E G(O,V) xt n-dimensional state
y k-dimensional output t
vt p-dimensional white noise
t E T-{O,l,...,t1} time index set
(1)
Compensator: ztfl - Ftzt } Gtyt
Assumptions: zt is an n-dimensional process,z~ - m
Filter problem. determine Ft, Gt, t E T such that zt is an unbiased, minimum variance estimator of xt.
nxn From the unbiasedness, we infer Ft - A-GtC. Now let et :- xt-zt and Et: T--}R be the variance of et, then it remains to determine Gt, t E T with respect to
a suitable criterion, implied by the minimum variance condition.
The resulting problem will be formulated as follows:
Given the matrix recursion for the variance of the error
~ttl - (A-GtC)Et(A-GtC)T t (M-GtN)V(M-GtN)T, ~G
(2)
and the terminal costs
E[ J ]- E[eT M e ]- trace M F , M ~ 0 ( 3)
t1 t1 t1 t1 t1 t1 t1
determine Gt, t E T to minimize E[Jt ] 1
This problem will be solved with the aid of the matrix minimum principle. App1i-cation of the MMP yiel.ds:
Proposition
3
-Pt - (A-GtC)T Pttl(A-GtC) P - M ~ 0
tl tl
and for the first-order condition (by differentiation of the hamiltonian H)
aG - Pttl [A Et CT t MVNT - Gt (C Et CT f NVNT) ]- Onxk t
Proof. Appendix A, AS, A6 and A7 and cf. I 3.2.10 and I 3.2.11.t
Remarks. For gr.~ater clearity zero matrices will be subscripted with their dimensions.
From (4): Pt - Pt, t E T.
Here we pause a moment to discuss the result. It is not straightforward to
con-clude immediately from (5) the desired result
Gt -(A Et CT f M~INT) (C Et CT t NVNT) -1 4 Gt
since (Gt-Gt) is premultiplied with Pr}~.
~ ~
From
Pttl(Gt-Gt) - Onxk we can only deduce Gt - Gt if Pt}1 is nonsingular. This is always the case in the continuous-time setting, but not in discrete-time. If,
for example, the state matrix A is singular, it may happen that (A-GtC) is singular and by (4) Pt, t ~ tl becomes singular. This will be illustrated in the
following example.
(4)
(5)
(6)
Example
Define the Gaussian syster~ 'xtfl - Axt f B w1t
yt - Cxt f w2t
(7)
where wlt and w2t are independent white noises.
Apply a transformation to bring (7) into standard form (1)
xtfl - Axt t[ B 0] (W1) xtfl - Axt f Mvt
' 2 or yt - Cxt f Nvt
yt - Cxt f[0 I] (W11 vt E G(O,V) 2
Let xt :- E[xtlFt-1] where Ft-1 ~ a{ys, 0 ~ s ~ t-1}, then the Kalman filter obeys
4 -xttl - Axt f Kt [yt - Cxt ] Kt -(A Et CT t MV:~IT) (C Et CT t NVNT) -1 ~tfl - A Et AT f MVMT - Kt(C Et AT t NVMT) , ~0
Now take n-2, k-1, P-3, B-V-E~-I, m0 -(~), C- (1 0)
MVNT - (~)
-a, E R, a~ 0 then MVMT - NVNT - 1 and
~ C 1 0
Note that the pair (A,C) is observable since CA - 1 a
has rank two if a~ 0.
For t- 0 we have: K- 1(1)0 2 0 ' A-K C0
F.1 -For t - 1: K1 aZ t 3~2 0 0 2a2f3 2 2a2t5 2a2t5 0 1 A-K1C i
and by induction it is easily seen that A-KtC is singular
for all t E T.
T
Hence Pt - ( A-KtC) Pt}1(A-KtC) will be singular for all t ~ tl.
Remark. At several occasions in the literature it is observed that in the dis-crete-time case the system matrix A is in fact the state transition matrix, and, by analogy of the continuous-time case, should be assumed nonsingular. By the same analogy Pt is assumed to be positive-definite.
-1 This enables to deal with Pt .
Now return to the first-order condition (5) and write it as
~
Pttl(Gt-Gt) - Onxk
-If we not assume that Pt, t E T is positive-definite, conclusions about Gt can only be made on the space spanned by the independent columns of Pt; on the
5
-space of Pt no conclusions can be made. It is still an open question whether there exists an alternative procedure to obtain an explicit expression for Gt, e.g. some recursive argument. Hoorever, the minimum principle only provides necessary conditions. Candidates for the optimal solution must be submitted to the sufficient conditions (i.e. the Hamilton-Jacobi-Bellman partial differential equation), which may in addition provide useful facts as uniqueness, global pro-perties, regularity conditions.
Evidently, a reasonable candidate is indeed Gt - Gt. And, as is well-known, from HJB-theory it can be proven that under certain conditions Gt - Gt is in-deed the unique solution for this problem.
The dual problem, i.e. the control problem, has been analyzed extensively in the vector-case. Then a full theory is availàble: HJB-theory provides sufficient conditions and the convexity of the Hamiltonian can be verified immediately,
2
since a2 can be calculated at once. au
3. The LQG- problem
3.1. The combined filter and control problem is dealt with in I 3.3. Just as in the filter case no constructive method can be found to obtain explicit expressions for the candidates of the optimal solution. But still reasonable candidates can be suggested, see the verification lemma below. The LQG-problem
is also discussed for the other representations of I 3.3.8, and since this ana-lysis is important, it will be presented in some detail.
First we regard the (é)-representation. Consult I 3.3 if necessary for details. To be complete, we repeat the equations for the system and the corresponding compensator.
P :
C .
(8)
(9)
In addition to the definitions of (1), ut is here the m-dimensional input or control process. Notice that the structure of the compensator is not as general as possible, but already appropriate
Let et :- xt - zt, then
E(2) .
for the problem to be discussed here.
A t BLt - BLt xt M t 0 A- KtC e t M- KtN tl-1
~~ o
J - (xTeT) ` ~ ( e) t f E SO 0 1 t-0 tfl - Axt f But f Mvt, x0[
yt - Cxt f Nvtt~-1 - Azt f But t Kt [Yt - Czt ] ut - Ltzt
i
(xTeT) i
` - LTRL
2nx2n
Let Et : T-~ R be the variance of (e) t, then:
-~-Now we can formulate the LQG-problem:
Given the "state-equation" ( 10), find Kt, Lt, t E T to minimize E[J].
Proposition. For (10) and (11), we obtain for the costate Pt: T-~R
~ T Pt - A Pttl A} Qt Qt 0 P - 1 tl 0 0
and for the two first-order conditions
[ BTP A f (BTP B t R) L](E - ET ) f
aH T 11 T11 t il 12
áLt - [ B P11BLt t B P12 (A-KtC) f RLt ](E22 - E12) - Omxn
áH P12 [(AfBLt) E12CT - BLtE22CT f MVNT] f aKt
-P22 [AE22CT f MVNT - Kt (CE22CT f NVNT) ]- Onxk
Proof. The costate equation follows from differentiating
H(Et' Ptfl' Kt, Lt) - tr{A Et A T Ptfl } MVM T Ptfl } Qt ~t}
2nx2n
(12)
(13)
(14)
with respect to Et; the first-order conditions follow from elaborating H in terms of ~T1 ~12 -: Et and pTl P12 -, pt}1 and
differentia-12 22 t 12 22
tfl ting with respect to Kt and Lt.
The arguments ttl and t of Pttl and Et resp. are omitted for brevity. ~ Verification lemma
The pair
Lt - - [BTP11B f R] -1 BTP11A
Kt -(AE22CT f MVNT) (CE22CT t NVNT)-1
satisfies the first-order conditions (13) and (14).
Proof. From (10) and (12) we have by simply expanding
~22(tfl) - (A-KC)E22(A-KC)T f (M-KN)V(M-KN)T pll (t) -(AfBL)T pll (t}1) (AfBL)
t LTRL f Q P12 (t) -(Ai-BL) T P12 (A-KC) -(AfBL) T P11BL - LTRL
p22 (t) - LTBTpi 1BL -(A-KC) T p12 BL - LTBT P12 (A-KC) f(A-KC) T P22 (A-KC) f LT RL
Hence
(~22 - ~12) tfl - (A}BL) (E22 - ~12) (A-KC) T
- Kt [C E22 AT f NVMT -(C E22 CT t NVNT) KT ] , ~nxn
P12 (t) -- Lt [BT pl l A }(BT pl l B t R) Lt ] t(A-KC) T P12 (tfl )( AfBL) , Onxn and the above choice of Kt and Lt sets:
for all t E T (~22 - ~12) t - ~nxn
T - p for all t E T p12(t) nxn
Now the two terms in (13) and (14) both vanish.
3.2. Along similar lines an equivalent result can be obtained for the (Z)-repre-sentation. The same notation is used as in 3.1.
xtf 1 ,,ztfl KtC BLt xt M f vt A t BLt - KtC zt KtN 'pt p tl-1 O E[J] - tr 1 Et f E tr ~ 0 0 1 t-0 ~tfl - A Et A~ f MVM , 1 LT ~Q 0~ costate : Pt - A T Pt} 1 A f Qt , t 1 0 0 first-order conditions: 1.0,~ (15) (16) atí - [BT(p11tP12)A t BT(p12tP22)KtC] ~12 } (17) aLt [RLt f BT (P11 } P12 } p12 f P22) BLt f BT (p12 t P22) (A-KtC) ]~22 - Omxn
óH p12[A(E11 - E12)CT f BLt(E12 - E22)CT f MVNT ] t
(18) óKt
Now the verification lemma is slightly more involved.
Verification lemma
Lt --[ BT (P11 ~- P12) B f R]-1 BT (P11 ~- P12) A The pair
Kt - IA(E11 - E12)CT t MVNT ] IC(E11 - E12)CT t NVNT ]-1
satisfies the first-order conditions (17) and (18).
Proof. From (15) and (16) we have
E 1 1( tf 1) - A E 1 1 AT t A E 12 LT BT f BL E 12 AT t BL E 2 2 LT BT t MVMT
E 12 ( tf 1) - A E 11 CT KT f A E 12 ( AfBL-KC )-T -t- BL E i 2 CT KT f BL E 2 2( AfBL-KC ) T t f MVNT KT
E22 (ttl) - K C E11 CT KT t K C E12 (AfBL-KC) T f (A~-BL-KC) E12 CT KT t t (AfBL-KC)E22 (AfBL-KC)T f KNVNT KT
P11(t) - ATP11AfCTKTP12AtATP12 KC t CTKTP22KCfQ
P12 (t) - AT P11 BL t CT KT P12 BL t AT P12 (AfBL-KC) f CT KT P22 (A-1-BL-KC)
Y221~1 - LT DT P11 liL ~(AIBL-KC)T P12 BL ~ LTBT P12(AtBL-KC) ~ f(AtBL-KC) T P22 (AfBL-KC) -F- LT RL
and therefore
(ET - F. ) - (AfBL-KC) (ET - E ) (A-KC) T f
12 22 ttl 12 22
Kt [C ( E 1 1- E 12 ) AT t NVMT -{NVNT ~- C( E 11 - E 12 ) CT }KT] , ~nxn (P12 f P22) t-(AfBL-KC) T(P12 f P22) (AfBL)
-1-Lt [BT (P11 f P12) (AtB-1-Lt) f R-1-Lt ]' ~nxn
Now the above choice of Kt and Lt makes E12 -~22 - Onxn and Pi2 } P22 - Onxn for all t E T and the two terms in (17) and in (18) both vanish. ~
Remark. Lt and Kt are driven by (P11 } P12)ttl and (E11- ~12)t resn. Using the expressions for Lt and Kt as given in the verification lemma, these recursions obey:
(P11 } P12) t- AT (P11 } P12) ttl (AfBLt) f Q
lU
-3.3. Summary
Since the notation E12, P22,... has been used throughout the different repre-sentations, the mnemonic Exz' ~ex' Pxx'" ' should be used now to compare the results of 3.1. and 3.2. Together with I 3.3.17 we have for the three different representations of the LQG-problem.
~~1) : (Z) representation -P - P xz zz ~~2) : (é) representation ~~3) : (é) representation EZe - 0 P - 0 ex P - P zz ze
11
-4. A remarkable formula for the LQG-case
Here we specialize to the (é)-representation. Analogous calculations can be made for the other two representatíons.
We repeat the model, costate equation and first-order conditions, see I 3.3.12-3.3.15. Let A M -A-KtC `S`t ut ~ K N' t K N t
then the state and costate equation respectively are
~tfl - AEtAT f MVMT
~ T ~
Pt - A Ptf 1 A } Qt '
The first-order conditions are:
( `1
`1 ~
Qtl
Qtl ~
3H ~(BTP11BfR)Lt t BTP11A]~11 }
iiLt -~BT pl l KtC f BT P12(A-KtC) ]~12 - OmXn
aH (P11 - P12) [(AfBLt) E12 CT f Kt(C E22 CT f NVNT)] ~ aKt -(p12 - p22) ~A ~22 CT f MVNT - Kt(C E22 CT f NVNT)] - Onxk
12
-[(AtBLt) E12 CT t Kt(C E22 CT t NVNT) ] Kt t ~11(ttl)
-(AtBLt) ~ il(AtBLt) T t Kt C E 1 2(Ai-BLt) T~~T T [A E22 CT t MVNT - Kt(C E22 CT t NVNT)] Kt t E (tfl)
-12
(A-Kt C) E12 (AtBLt) T. ~nxn
of Lt and Kt
The E- and P-blocks at the RHS of these expressions should be evaluated at t and ttl respectively.
Now we see the remarkable fact that the coefficients show up in the first-order conditions (21) and (22).
nx2n 2nxn
of IIt : T-~ R and St : T-~ R such that
~t '- [ (P11 - P12) (P12 - P22)] t
(25)
(26)
in (23)-(26), This leads to the definition
If we postmultiply (23) and (24) by E11(t) and E12(t) resp. and premultiply (25) and (26) by (P11 - P12) ttl and (P12 - P22) ttl resp. , after substitutíng the first-order conditions (21) and (22) and arranging the different terms, we arrive at
T[t St -(A-Kt C) T~tt1 A St ~ttl Stfl - ~tt1 A St ( A-F-BLt)
T
(27)
(28)
(27) can be seen as a backward recursion in Iit, (28) can be seen as a forward recursion in St.
Note at the dominant role of the product lIt}1 A St, which will show up again later in the Nash-compensator case and also in an alternative derivation of the LQG-result in section 5.
Also notice that from the optimal LQG-model we have P11 - P12 - 0 and E12 - 0 for all t E T, implying that T[t St - 0 for all t E T.
However, there seems to be no constructive way to obtain this or a símilar result (e,g. Ilttl A St - 0) directly from the first-order conditions combined with (23)-(26). It is conjectured that this result (]?tSt - 0 or ~ttl A St - 0) is intimately related with the separation property.
Therefore the separation property must be understood in the LQG-case and to that aim a more concise and insightful derivation of the LQG-result will be presented
14
-5. The separation property for the LQG-model
In this section we use the model of section four, i.e. the (é)-representation, under the additional assumption that system- and measurement noise are uncorre-lated (MVNT - 0). This is no serious restriction, but will elucidate the duality between the control and the filter part more clearly.
The presented derivation here will be different from the one in section 3. Ex-panding the state and costate equations is rather tedious, especially in the
two-DM case with dimensions 3nx 3n and nine blocks for Pt and Et. This can be E R2nX2n
such that avoíded by introducing E1, E2
Ei ;- (ó) E2 :- (~) where I is the n x n unity matrix.
Now, cf. (19) and (20) the augmented system is characterized through:
A - E1 A E1 t E2 A E2 f E1 BL E1 t(E1-E2) KC E2
,~, T (E1-E2)KNVNT KT(Ei E2) T t E2 MVMT E2 f
Q MVM -E2 MVNT KT (E1--E2) T t(E1--E2) T KNVMT -E2 - (E1fE2)Q(E1fE2)T t E1 LT RLE1 H - tr {A Et ATPttl } MVM T Pt}1 } Qt ~t} aH - BT ET P A E E -f- RL ET E E - 0 (29) aLt 1 ttl t 1 t 1 t 1 mxn aH -(E -E ) T p A E E CT f( E -E ) T P (E -E ) K NVNT - 0 (30) aK 1 2 tfl t 2 1 2 tfl 1 2 t nxk t
By expanding (29) and (30) we obtain the earlier derived expressions (21) and (22). Notice the duality between (29) and (30). Obviously the factor Pt}lA ~t is invariant under duality: Now from this viewpoint we will investigate the se-paration property. To obtain a separation result, we need a solution of (29) and
(30) in the form:
K - K(E)
15 -T
So from ( 29) it is seen that ElEt E1 must cancel,
from (30) it is sec:n that (E1-E2)T Pttl ( E1 -E2) must cancel.
This is indeed the case, as will be verified now.
We know that for the optimal solution of the LQG-model we have
L
T
E12 - 0
T
P11-P12 - 0
Using this condition, the first-order conditions reduce to:
AfBL K C ~E i) BTE1Ptf1A~tE1 (31)
BT(p11P12)
t t 11
0 A-Kt C ~0
-(p12-P22) (A-KC) E22 C -(E1-E2)Ptfl E2 (A-KC) E2 Et E2 C AtBL K C 0 ii) (E1-E2) T ptfl A~t E2 CT (31) (0 P12 p22) t t CT 0 A-Kt C E2` T - T T So what happens?
If we insert (31) in the áL - 0 equation, A"reduces" to t E1 (AfBLt) E1 -A for all t E T AfBLt BT P11 (AfBLt) E11 - BT E1 Pttl E1 (AtBLt) E1 Et E1
and in the áK - 0 equation t
` 0 0
"reduces" to EZ(A-Kt C) E2 - 1 ~ So, from (29) ,(30) and (31) we conclude
BT E1 pttl E1 (AfBLt) E1 Et E1 f RLt Ei Et E1 - Omxn
(E1-E2)
T Ptfl (E1E2) (AKt C) E2 Et E2 CT
-(E1-E2)T pttl (E1-E2) Kt NVNT - Onxk
(31)
Assuming Ei Et E1 and
16
-BT E1 Ptfl E1 (AfBLt) f RLt - Omxn
17
-Appendix A: The matrix minimum principle
Theorem
Given:
(A1) state equation Xt}1- Xt - F(t,Xt,Ut), X~ tl-1
(A2) costs J- K(Xt ) f E L(t,Xt,Ut)
1 t-0
(A3) Hamiltonian H(Xt'Pttl'Ut) ~ L(t,Xt,Ut) f tr [F (t,Xt,Ut) Ptfl~ where X : T ~ Rnlxn2 mlxm2 U : T ; R nixn2 mlxm2 nlxn2 F: T x R x R -~ R nlxn2 K : R -~ R nlxn2 mixm2 L: T x R x R -~ R P : T x Rnlxn2 nlxn2 nlxn2 mlxm2 H: R x R x R ~ R
T - {O,l,...,ti} time index set.
If Ut is the optimal unconstrained control and Xt the corresponding state trajec-tory, then there exists a costate matrix Pt, t E T such that
18
-(A7 ) aH
aut
- 0x mlxm2
19
-Appendix B: Reduced order compensators
A still partially unsolved problem is the reduced-order compensator. xtf i- A xt t B ut t r~ vt
Let the system be:
yt - C xt f N vt
and let the compensator have the most general structure zttl - Ft zt t Gt ut f Kt yt
ut - Lt zt
where xt is the n-dimensional state process,
zt is the nl (~n) dimensional compensator and Ft, Gt, Kt, Lt are unknown matrices.
(B1)
(B2)
These four unknowns cannot be determined in a straightforward way if nl ~ n; compare part I, section 4.
Cohen [1] gives only necessarily conditions for this problem, derived for a
COT1tir1Lni~c-tim2 modcl ir. an approach compïeteiy parallel as presented here.
Sims [4] assumes that the matrix F is known by some device and applies the MMP
to determine Kt (no control process present).
Due to these problems we will restríct to n-dimensional compensators.
-20-References
[1] G. Cohen: Partage de 1'information entre systèmes stochastiques a priori indépendants, RAIRO Automatique~Systems Analysis and Control, vol. 15, pp. 45-68 (1981).
(2] M. Athans, Optimal Control, McGraw-Hill (1966). P.L. Falb:
[3] M.D. Merbis: On the compensator.
Part I: problem formulation and preliminaries, Reeks "Ter Discussie", 82.12, July 1982.
1
IN 1982 REEDS V~RSCFiENEN:
O1. W, van Groenendaal Building and analyzing an jan,
econometric model with the use of a hybrid computer; part I.
02. M.D. Merbis System properties of the jan,
interplay model
-03. F. Boekema Decentralisatie en regionaal maart
sociaal-economisch beleid
04. P.T.W.M. Veugelers Een monetaristisch model voor
maart de Nederlandse economie
O5. F. Boekema Morfologie van de "Wolstad".
april Over het ontstaan en de
ont-. wikkeling van de ruimtelijke
geleding en struktuur van Tilburg .
06. P. van Geel Over de (on)mogelijkheden mei
van het model van Knoester. 07, J.H.M. Donaers,
F.A,M, van der Reep
08. R.M.J. Heuts
09. B.B. van der Genugten
10. J. Roemen
li. J. Roemen
12. M.D. Merbis
13. P. Slangen
14. M.D. Merbis
De betekenis van het monetaire beleid voor de Nederlandse eco-nomie, presentatie van een ana-lys~ aan de iiand van een een-voudig model
The use of non-linear
trans-formation in ARIMA-Models when the data are non-Gaussian distributed
mei
juni Asymptotic normality of least
squares estimators in auto-regressive linear regression
models . j ~i
Van koetjes en kalfjes I juli
Van koetjes en kalfjes II juli On the compensator
Part I
Problem formulation and
prelimi-naries juli
Bepaling van de optimale beleids-parameters voor een stochastisch kasbeheersprobleem met continue
controle aug.
Linear - Quadratic - Gaussian
2
15. P. Hinssen J. Kriens ~
J. Th. van Lieshout
Een kasbeheermodel onder
onzekerheid sept.
16. A. Hendriks en
T, van der Bij-Veenstra
17. F.W.M. Boekema A.J. Hendriks L.H.J. Verhoef 18. B. Kaper 19. P.F.P.M. Nederstigt 20. J.J.A. tioors 21. J. Plasmans H. Meersman 22. J. Plasmans H. Meersman
23. B.B. van der Genugten
24. F.A. Kense
25. R.T.P. Wiche
26. J.A.M. Oonincx
"Van Bedrijfsverzamelgebouw
naar Bedrijvencentrum" okt. Industriepolitiek, Regionaal
beleid en Innovatie okt.
Stability of a discrete-time, macroeconomic disequilibrium model.
Over de toepasbaarheid van het Amerikaanse 'Diagnosis Related Group'-systeem in Nederland
Auditing and Bayes' Estimation
An Econometric Quantity Ratio-ning Model for the Labour Market.
okt.
nov. nov.
nov. Theorieën van de
werkloos-heid. nov.
Een model ter beschrijving van de ontwikkelino van de veestapel
in Nederland. nov.
De omzet~artikel
concentratie-curve als beleidsinstrument nov. Populaire wetten~specificatieve
wetten, oftewel
IN 1983 REEDS VERSCHENEN O1. F. Boekema L. Verhoef 02. R.H. Veenstra J. Kriens 03. J. Kriens J.Th. van Lieshout J. Roemen P. Verheyen 04. P. Meys 05. H.J. Klok 06. J. Glombowski M. Kruger 07. G.J.C.TH. va-, Schijndel 08. F. Boekema L. Verhoef Enterprise Zones.
Vormen Dereauleringszones een adeauaat instrument van
regio-naal sociaal-economisch beleid? jan. Statistical Sampling in Internal Control Systems by Using the A.O.Q.L.-System.
Management Accounting and Operational Research
jan.
jan.
Het autoritair etatisme jan.
De klassieke politieke
economie geherwaardeerd febr.
Unemployment benefits and
Goodwin's growth cycle model febr. Inkomstenbelasting in een
dynamisch model van de onder-neming
Local initiatives: local enter-prise agency~trust, business in
the community
febr.